Unmanned aerial vehicle auxiliary edge calculation method and device based on heterogeneous hardware and product

Abstract

The invention relates to the technical field of mobile edge computing, in particular to an unmanned aerial vehicle auxiliary edge computing method and device based on heterogeneous hardware and a product, and the method comprises the steps: obtaining first time delay and first energy consumption when a terminal device processes a target task; acquiring second time delay and second energy consumption when the unmanned aerial vehicle processes the target task; acquiring third time delay and third energy consumption when the edge cloud server processes the target task; obtaining an unloading decision and a resource allocation strategy of the target task based on the expression of the total processing time delay and the expression of the total energy consumption; and processing the target task by the terminal device, the unmanned aerial vehicle or the edge cloud server based on the unloading decision and the resource allocation strategy. By deploying heterogeneous hardware on the unmanned aerial vehicle and the cloud edge server, staged target task processing is realized, the advantages of each processing unit can be brought into full play, efficient distribution and accelerated processing are realized, the overall performance and efficiency are improved, and delay and energy consumption are reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of mobile edge computing technology, and in particular to a drone-assisted edge computing method, device, and product based on heterogeneous hardware. Background Art

[0002] With the development of 5G technology and the large-scale commercial use of artificial intelligence (AI), user demand for computing continues to grow. Emerging compute-intensive and latency-sensitive applications (such as virtual reality, augmented reality, autonomous driving, and intelligent interaction) are unable to provide a high-quality user experience due to the distance to cloud computing centers and local computing resources. To address this, the concept of Multi-Access Edge Computing (MEC) has emerged. Using drone platforms equipped with edge computing servers to provide users with drone-assisted edge computing services is becoming a leading-edge solution that is attracting considerable attention.

[0003] However, existing drone-assisted edge computing approaches rely solely on central processing units (CPUs), neglecting the combined use of CPUs, graphics processing units (GPUs), and data processing units (DPUs), lacking research on heterogeneous hardware computing. Given the massive data explosion, increasingly intensive computing workloads, and increasingly complex computational tasks in the Industrial Internet of Things (IIoT) environment, this traditional CPU-centric architecture is increasingly unable to meet the computing power and performance requirements. Summary of the Invention

[0004] In view of the above problems, the present disclosure is proposed, and the present disclosure provides a drone-assisted edge computing method, device and product based on heterogeneous hardware.

[0005] According to one aspect of the present disclosure, a UAV-assisted edge computing method based on heterogeneous hardware is provided, comprising:

[0006] Obtaining a first time delay and a first energy consumption when a target task is processed by a terminal device; and, based on heterogeneous hardware deployed on the drone, processing the target task in stages; obtaining a second time delay and a second energy consumption when the target task is processed by the drone based on the time delay and energy consumption required for processing the target task at each stage; and, based on the heterogeneous hardware deployed on the edge cloud server, processing the target task in stages; and obtaining a third time delay and a third energy consumption when the target task is processed by the edge cloud server based on the time delay and energy consumption required for processing the target task at each stage;

[0007] Based on the first delay, the second delay, and the third delay, an expression for the total processing delay of the target task is obtained; and based on the first energy consumption, the second energy consumption, and the third energy consumption, an expression for the total energy consumption of the target task is obtained;

[0008] Based on the expression of the total processing delay and the expression of the total energy consumption, an offloading decision and a resource allocation strategy for the target task are obtained; the offloading decision is used to indicate that the target task is processed by the terminal device, the drone, or the edge cloud server, and the resource allocation strategy is used to indicate the computing resources of the heterogeneous hardware to be allocated when the target task is processed in stages;

[0009] Based on the offloading decision and resource allocation strategy, the target task is processed by the terminal device, drone or edge cloud server.

[0010] Furthermore, according to an aspect of the present disclosure, a drone-assisted edge computing method based on heterogeneous hardware includes a central processing unit, a graphics processing unit, and a data processing unit;

[0011] Based on the heterogeneous hardware deployed on the drone, the target task is processed in stages, including:

[0012] Based on a data processing unit deployed on the UAV, the target task is offloaded to the UAV and the target task is decrypted;

[0013] Based on a central processing unit deployed on the UAV, the decrypted target task is decomposed into parallel subtasks and serial subtasks;

[0014] Performing hybrid computing by the central processing unit and the graphics processing unit, wherein the central processing unit computes serial subtasks and the graphics processing unit computes parallel subtasks;

[0015] The calculation results of the serial subtasks and the parallel subtasks are aggregated by the central processing unit to obtain the average calculation amount of the central processing unit, the graphics processing unit and the data processing unit respectively.

[0016] In addition, the heterogeneous hardware-based drone-assisted edge computing method according to one aspect of the present disclosure further includes, before obtaining the first latency and the first energy consumption when the terminal device processes the target task:

[0017] A system model of drone-assisted edge computing is constructed, which includes a terminal device, a drone and an edge cloud server; wherein, a central processing unit is deployed on the terminal device; heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the drone; and heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the edge cloud server.

[0018] In addition, according to an aspect of the present disclosure, a UAV-assisted edge computing method based on heterogeneous hardware is used to obtain a first latency and a first energy consumption when a terminal device processes a target task, including:

[0019] Obtaining a first latency based on computing resources of a central processing unit of the terminal device allocated to the target task and an average computing load of the central processing units deployed on the terminal device;

[0020] A first energy consumption is obtained based on the first time delay.

[0021] In addition, according to an aspect of the present disclosure, the heterogeneous hardware-based drone-assisted edge computing method obtains a second delay and a second energy consumption when the drone processes the target task based on the delay and energy consumption required for processing the target task at each stage, including:

[0022] Calculate the transmission delay for offloading the target task to the UAV based on the data transmission rate of the uplink computing task from the terminal device to the UAV;

[0023] Calculating a decryption delay required to decrypt the target task based on computing resources of a data processing unit allocated by the drone to the target task and an average computing workload of the data processing unit;

[0024] Calculating a two-stage delay required for decomposing and aggregating the target task based on computing resources of a central processing unit of the drone allocated to the target task and an average computing load of the central processing unit;

[0025] Calculating the computational latency of the serial subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the central processing unit of the drone allocated to the target task, and the average computational load of the central processing unit; and calculating the computational latency of the parallel subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the graphics processing unit of the drone allocated to the target task, and the average computational load of the graphics processing unit;

[0026] Obtaining a hybrid computing delay required for hybrid computing based on the computing delay of the serial subtask and the computing delay of the parallel subtask;

[0027] Based on the transmission delay, decryption delay, two-stage delay and hybrid computing delay, a second delay and a second energy consumption when the drone processes the target task are obtained.

[0028] In addition, according to an aspect of the present disclosure, the heterogeneous hardware-based drone-assisted edge computing method obtains the offloading decision and resource allocation strategy of the target task based on the expression of the total processing delay and the expression of the total energy consumption, including:

[0029] Based on the expression of the total processing delay and the expression of the total energy consumption, adjusting the weight coefficients of the total processing delay and the total energy consumption, minimizing the total processing delay and the total energy consumption, and obtaining a utility function of the weighted sum of the total processing delay and the total energy consumption;

[0030] The utility function is solved using a two-level optimization algorithm to obtain the offloading decision and resource allocation strategy of the target task.

[0031] In addition, according to an aspect of the present disclosure, the UAV-assisted edge computing method based on heterogeneous hardware uses a two-layer optimization algorithm to solve the utility function and obtain the offloading decision and resource allocation strategy of the target task, including:

[0032] Inputting the number of drones and the location coordinates of the terminal devices into the self-organizing map neural network model to obtain the deployment location of each drone;

[0033] Based on the deployment position of each UAV, a hybrid chaos local search PID search algorithm is used to solve the utility function to obtain the offloading decision and resource allocation strategy of the target task.

[0034] In addition, according to the heterogeneous hardware-based drone-assisted edge computing method of one aspect of the present disclosure, the offloading decision includes an offloading factor and an offloading mode, and the values ​​of the offloading factor and the offloading mode are 0 or 1;

[0035] Based on the offloading decision and resource allocation strategy, the terminal device, the drone, or the edge cloud server processes the target task, including:

[0036] When the offloading factor is equal to 0, based on the resource allocation strategy, the central processing unit deployed on the terminal device processes the target task;

[0037] When the offloading factor is equal to 1 and the offloading mode is equal to 1, the target task is offloaded from the terminal device to the UAV, and the target task is processed in stages based on the heterogeneous hardware deployed on the UAV and the resource allocation strategy;

[0038] When the offloading factor is equal to 1 and the offloading mode is equal to 0, the target task is offloaded from the terminal device to the edge cloud server through the drone, and the target task is processed in stages based on the heterogeneous hardware deployed on the edge cloud server and the resource allocation strategy.

[0039] According to another aspect of the present disclosure, a UAV-assisted edge computing device based on heterogeneous hardware is provided, comprising:

[0040] The first computing module is configured to obtain a first time delay and a first energy consumption when a target task is processed by a terminal device; and, based on the heterogeneous hardware deployed on the drone, process the target task in stages; and, based on the time delay and energy consumption required for processing the target task at each stage, obtain a second time delay and a second energy consumption when the target task is processed by the drone; and, based on the heterogeneous hardware deployed on the edge cloud server, process the target task in stages; and, based on the time delay and energy consumption required for processing the target task at each stage, obtain a third time delay and a third energy consumption when the target task is processed by the edge cloud server;

[0041] The second calculation module is used to obtain the total processing delay (t m ) expression; and, based on the first energy consumption, the second energy consumption and the third energy consumption, obtain the total energy consumption (e m )

[0042] an acquisition module, configured to acquire an offloading decision and a resource allocation strategy for the target task based on an expression for the total processing delay and an expression for the total energy consumption; the offloading decision being used to indicate that the target task is processed by the terminal device, the drone, or the edge cloud server, and the resource allocation strategy being used to indicate computing resources of heterogeneous hardware to be allocated when processing the target task in stages;

[0043] A task execution module is used to process the target task by the terminal device, drone or edge cloud server based on the offloading decision and resource allocation strategy.

[0044] According to another aspect of the present disclosure, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method of the above aspect.

[0045] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method of the above aspect is implemented.

[0046] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements the method of the above aspect when executed by a processor.

[0047] As will be described in detail below, according to a heterogeneous hardware-based drone-assisted edge computing method, device and product of the embodiments of the present disclosure, by introducing the concept of heterogeneous hardware, heterogeneous hardware integrating a central processing unit, a graphics processing unit and a data processing unit is deployed on the drone and the cloud edge server, thereby realizing phased processing of target tasks, fully leveraging the advantages of each processing unit, and realizing efficient allocation and accelerated processing of target tasks. Compared with the traditional computing architecture that only uses a single type of processing unit, the computing architecture using heterogeneous hardware can significantly improve the overall performance and efficiency of task processing, and reduce the average task delay and energy consumption.

[0048] It is to be understood that both the foregoing general description and the following detailed description are exemplary, and are intended to provide further explanation of the technology as claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other purposes, features, and advantages of the present disclosure will become more apparent through a more detailed description of the embodiments of the present disclosure in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure and are not intended to limit the present disclosure. In the drawings, the same reference numerals generally represent the same components or steps.

[0050] Figure 1 is a flowchart illustrating the application of a heterogeneous hardware-based drone-assisted edge computing method according to an embodiment of the present disclosure.

[0051] Figure 2 It is a model architecture diagram illustrating the application of a heterogeneous hardware-based drone-assisted edge computing system according to an embodiment of the present disclosure.

[0052] Figure 3 is an architectural diagram illustrating the application of a four-stage task processing model according to an embodiment of the present disclosure.

[0053] Figure 4 is another flowchart illustrating the application of the heterogeneous hardware-based drone-assisted edge computing method according to an embodiment of the present disclosure.

[0054] Figure 5 is a framework diagram illustrating the application of the two-layer optimization algorithm ToSMPSA according to an embodiment of the present disclosure.

[0055] Figure 6 2 is a diagram illustrating the area division and drone deployment location diagram according to an embodiment of the present disclosure.

[0056] Figure 7 2 is a schematic diagram illustrating the structure of a UAV-assisted edge computing device based on heterogeneous hardware according to an embodiment of the present disclosure.

[0057] Figure 8 FIG2 is a schematic diagram illustrating the structure of a computer device according to an embodiment of the present disclosure.

[0058] Figure 9 is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present disclosure more apparent, the following will describe in detail exemplary embodiments of the present disclosure with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments of the present disclosure, and it should be understood that the present disclosure is not limited to the exemplary embodiments described herein.

[0060] With the development of 5G technology and the large-scale commercial use of artificial intelligence (AI), user demand for computing continues to grow. Emerging compute-intensive and latency-sensitive applications (such as virtual reality, augmented reality, autonomous driving, and intelligent interaction) are unable to provide a high-quality user experience due to the distance to cloud computing centers and local computing resources. To address this, the concept of Multi-Access Edge Computing (MEC) has emerged. Using drone platforms equipped with edge computing servers to provide computing services to users is becoming a leading-edge solution that is attracting considerable attention.

[0061] Given the massive data explosion, increasingly intensive computing workloads, and increasingly complex computing tasks in the Industrial Internet of Things (IIoT) environment, traditional central processing unit (CPU)-centric data center architectures are increasingly facing performance bottlenecks. Consequently, leading-edge devices are increasingly relying on heterogeneous computing platforms that integrate different types of processing units to meet the requirements of latency-sensitive and compute-intensive tasks. Data-centric computing architectures built around data processing unit (DPU) systems are becoming the trend of evolution. These architectures integrate three different types of processors: central processing units (CPUs), graphics processing units (GPUs), and data processing units (DPUs), aiming to leverage the strengths of each to improve data processing efficiency and overall system performance.

[0062] In multi-access edge computing systems, the use of drone platforms equipped with edge computing servers to provide computing services to users is becoming a highly sought-after cutting-edge solution. This approach leverages the high flexibility and maneuverability of drones, enabling rapid deployment and mobility without the constraints of fixed infrastructure. Compared to traditional fixed edge computing nodes, drone platforms can dynamically adjust their position based on user needs and geographic location, maximizing proximity to data sources. This significantly reduces data transmission latency and network congestion, and improves the responsiveness of computing services.

[0063] However, existing drone-assisted edge computing approaches only consider optimal task offloading strategies for drones under energy-constrained conditions. Target tasks can only be processed locally on terminal devices or offloaded to drones, employing a two-tier offloading model: terminal device-drone. These approaches fail to consider drones acting as aerial relays, forwarding target tasks from terminal devices to ground-based cloud edge servers, which can easily lead to drone overload. Furthermore, these modeling approaches focus solely on CPU processing for cloud edge servers, without considering the task processing characteristics of heterogeneous hardware service clusters.

[0064] The above describes, with reference to the accompanying drawings, a drone-assisted edge computing method, device, and product based on heterogeneous hardware according to an embodiment of the present disclosure. By introducing the concept of heterogeneous hardware, heterogeneous hardware integrating a central processing unit, a graphics processing unit, and a data processing unit is deployed on the drone and the cloud edge server, thereby realizing phased processing of target tasks, fully leveraging the advantages of each processing unit, and realizing efficient allocation and accelerated processing of target tasks. Compared with the traditional computing architecture that only uses a single type of processing unit, the computing architecture using heterogeneous hardware can significantly improve the overall performance and efficiency of task processing, and reduce the average task delay and energy consumption.

[0065] The design comprehensively considers the characteristics of drone communication channels and the dual functions of drones as edge servers or aerial repeaters, fully utilizing the flexibility and maneuverability of drones, further optimizing the utilization of channel resources, and improving the robustness of the system and user experience in complex environments.

[0066] Heterogeneous hardware service clusters are deployed on drones and cloud edge servers to assist in the edge computing network. A four-stage heterogeneous hardware computing model is proposed to describe the computational process of the target task. The data processing unit (DPU) can save computing resources from the central processing unit (CPU) and improve overall computing power by flexibly offloading services such as virtualization, networking, storage, and security. Meanwhile, the graphics processing unit (GPU) accelerates graphics rendering and parallel computing tasks, adopting a massively parallel computing architecture to support high-throughput data processing, improving computing density and energy efficiency. Heterogeneous hardware achieves an overall improvement in computing performance and power efficiency through the coordinated operation of different types of processing units.

[0067] A two-layer optimization algorithm (ToSMPSA) was proposed, combining a self-organizing map neural network (SOM) with a hybrid chaos local search PID search algorithm (CLSPSA), to effectively solve this mixed-integer nonlinear optimization problem. Simulation results also show that compared with existing optimization algorithms, the ToSMPSA algorithm significantly improves solution accuracy and convergence speed, enabling faster optimal solutions. This allows for better support for complex scenarios such as drone collaboration and heterogeneous resource allocation, effectively improving the computational and energy efficiency of edge computing networks and significantly improving the overall performance of the system.

[0068] To facilitate understanding of this embodiment, a UAV-assisted edge computing method based on heterogeneous hardware disclosed in an embodiment of the present disclosure is first introduced in detail. The execution subject of the UAV-assisted edge computing method based on heterogeneous hardware provided in the embodiment of the present disclosure is generally a computer device with certain computing capabilities. The computer device includes, for example: a terminal device or a server or other processing device. The terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the UAV-assisted edge computing method based on heterogeneous hardware can be implemented by a processor calling computer-readable instructions stored in a memory.

[0069] like Figure 1 FIG. 1 is a flowchart of a UAV-assisted edge computing method based on heterogeneous hardware provided by an embodiment of the present disclosure, wherein the method includes S101 to S104:

[0070] S101: Obtain a first delay and a first energy consumption when a target task is processed by a terminal device; and, based on the heterogeneous hardware deployed on the drone, process the target task in stages; based on the delay and energy consumption required for processing the target task in each stage, obtain a second delay and a second energy consumption when the target task is processed by the drone; and, based on the heterogeneous hardware deployed on the edge cloud server, process the target task in stages; based on the delay and energy consumption required for processing the target task in each stage, obtain a third delay and a third energy consumption when the target task is processed by the edge cloud server.

[0071] Optionally, before S101, build a system model for drone-assisted edge computing, such as Figure 2 As shown, the system model includes a terminal device, a drone and an edge cloud server; wherein, a central processing unit is deployed on the terminal device; heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the drone; and heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the edge cloud server.

[0072] In one embodiment, if Figure 3 As shown in the figure, based on the heterogeneous hardware deployed on the UAV, the target task is processed in stages, including the following steps:

[0073] 1) Based on the data processing unit deployed on the UAV, offload the target task to the UAV and decrypt the target task;

[0074] 2) Based on the central processing unit deployed on the UAV, decompose the decrypted target task into parallel subtasks and serial subtasks;

[0075] 3) performing hybrid computing by the central processing unit and the graphics processing unit, wherein the central processing unit computes serial subtasks and the graphics processing unit computes parallel subtasks;

[0076] 4) aggregating the computation results of the serial subtasks and the parallel subtasks through the central processing unit to obtain the average computational load of the central processing unit, the graphics processing unit, and the data processing unit, respectively.

[0077] Similarly, the steps for staged processing of target tasks based on heterogeneous hardware deployed on cloud edge servers are the same as those for drones and will not be repeated here.

[0078] Optional calculation of expressions related to latency and energy consumption, including:

[0079] (1) When the terminal device processes the target task:

[0080] Obtaining a first latency based on computing resources of a central processing unit of the terminal device allocated to the target task and an average computing load of the central processing units deployed on the terminal device;

[0081] A first energy consumption is obtained based on the first time delay.

[0082] (2) When the drone is processing the target mission:

[0083] Transmission delay: Based on the data transmission rate of the uplink computing task from the terminal device to the drone, the transmission delay of offloading the target task to the drone is calculated.

[0084] Decryption delay: Calculate the decryption delay required to decrypt the target task based on the computing resources of the data processing unit allocated to the target task by the drone and the average computing workload of the data processing unit;

[0085] Two-stage delay: Based on the computing resources of the central processing unit of the UAV allocated to the target task and the average computing capacity of the central processing unit, the two-stage delay required for decomposing and aggregating the target task is calculated;

[0086] Hybrid computing latency: Calculate the computational latency of the serial subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computing, the computational resources of the central processing unit of the drone allocated to the target task, and the average computational workload of the central processing unit; and calculate the computational latency of the parallel subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computing, the computational resources of the graphics processing unit of the drone allocated to the target task, and the average computational workload of the graphics processing unit;

[0087] Obtaining a hybrid computing delay required for hybrid computing based on the computing delay of the serial subtask and the computing delay of the parallel subtask;

[0088] Based on the transmission delay, decryption delay, two-stage delay and hybrid computing delay, a second delay and a second energy consumption when the drone processes the target task are obtained.

[0089] (3) When the cloud edge server processes the target task:

[0090] Drones act as aerial relays, offloading target tasks from terminal devices to cloud edge servers;

[0091] Calculate the transmission delay of offloading the target task to the cloud edge server based on the data transmission rate of the uplink computing task from the drone to the cloud edge server;

[0092] Calculating a decryption delay required to decrypt the target task based on the computing resources of the data processing unit allocated by the cloud edge server to the target task and the average computing amount of the data processing unit;

[0093] Calculate the two-stage delay required for decomposing and aggregating the target task based on the computing resources of the central processing unit allocated to the target task by the cloud edge server and the average computing amount of the central processing unit;

[0094] Calculating the computational latency of the serial subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the central processing unit of the cloud edge server allocated to the target task, and the average computational load of the central processing unit; and calculating the computational latency of the parallel subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the graphics processing unit of the drone allocated to the target task, and the average computational load of the graphics processing unit;

[0095] Obtaining a hybrid computing delay required for hybrid computing based on the computing delay of the serial subtask and the computing delay of the parallel subtask;

[0096] Based on the transmission delay, decryption delay, two-stage delay and hybrid computing delay, a third delay and a third energy consumption when the cloud edge server processes the target task are obtained.

[0097] Specifically, regarding the data transmission rate: considering the characteristics of the communication channel of the UAV, this embodiment assumes that the total bandwidth is divided into two main parts: the sub-channel bandwidth B of the G2A link (Ground-to-Air Communication) G , A2N link (Air-to-Network Communication) sub-channel bandwidth B A To simplify the analysis, the location of the cloud edge server BS is (0,0,0), and the UAV U n The position of (x n ,y n ,h), the drone is suspended at a certain height h, the terminal device D m The position of (x m ,y m ).

[0098] (1) For G2A links: Due to the many different objects in the real environment acting as scatterers or obstacles, the radio signals transmitted by ground equipment or UAVs do not propagate in free space, but may be shadowed or scattered by man-made structures, resulting in additional path loss. Therefore, the simplified free space path loss (FSPL) model is not sufficient to accurately describe the communication between ground equipment and UAVs. On this basis, this embodiment uses a probabilistic path loss model that considers the probability of occurrence and path loss of LoS and Non-LoS (NLOS) communications to model the G2A link, providing a comprehensive description of G2A communication.

[0099] Finally, the data transmission rate of the uplink computing task from the terminal device to the drone is:

[0100]

[0101] in, Indicates terminal device D m and UAV U n The channel gain between Indicates drone U n The interference power signal of other devices in the coverage area, N0 is the noise power, p m Indicates terminal device D m The transmission power, B G Indicates the subchannel bandwidth of the G2A link.

[0102] (2) For the A2N link: Considering that the UAV is hovering at high altitude, the antenna reception of the cloud edge server BS is unobstructed, and the communication of the A2N link is mainly dominated by the LOS link. Therefore, the communication environment from the UAV to the edge server can be approximated as free space, and the FSPL model can be applied to A2N communication.

[0103] Finally, the data transmission rate of the uplink computing task from the drone to the cloud edge server is:

[0104]

[0105] in, Indicates drone U n The channel gain between the cloud edge server BS, p n Indicates drone U n The transmission power, B A Indicates the sub-channel bandwidth of the A2N link.

[0106] S102: Obtain an expression for the total processing delay of the target task based on the first delay, the second delay, and the third delay; and obtain an expression for the total energy consumption of the target task based on the first energy consumption, the second energy consumption, and the third energy consumption.

[0107] Specifically, the total processing delay is expressed as:

[0108]

[0109] Among them, t m represents the total processing delay, represents the first delay, represents the second delay, represents the third delay, α m represents the unloading factor, Indicates the uninstallation method.

[0110] The expression of total energy consumption is:

[0111]

[0112] Among them, e m represents the total energy consumption, represents the first energy consumption, represents the second energy consumption, represents the third energy consumption, α m represents the unloading factor, Indicates the uninstallation method.

[0113] S103: Based on the expression of the total processing delay and the expression of the total energy consumption, obtain the offloading decision and resource allocation strategy of the target task.

[0114] The offloading decision is used to indicate whether the target task is to be processed by a terminal device, a drone, or an edge cloud server. The resource allocation strategy is used to indicate the computing resources of heterogeneous hardware to be allocated when processing the target task in stages.

[0115] Specifically, S103 includes the following steps 1-2:

[0116] Step 1: Based on the expression of the total processing delay and the expression of the total energy consumption, adjust the weight coefficients of the total processing delay and the total energy consumption, minimize the total processing delay and the total energy consumption, and obtain a utility function of the weighted sum of the total processing delay and the total energy consumption. The utility function is as follows:

[0117]

[0118] Among them, ω t The weight coefficient representing the impact of total processing delay on terminal device performance, ω e The weight coefficient representing the impact of total energy consumption on the performance of terminal equipment, t m represents the total processing delay, e m represents the total energy consumption, α m represents the unloading factor, Indicates the uninstall method. Represent the resource allocation strategies of the central processing unit, data processing unit and graphics processing unit respectively, {x n ,y n} indicates the deployment location of the drone.

[0119] The utility function involves the optimization of six variables, namely α, β, k C ,k D ,k G ,{x n ,y n}, C1 and C2 determine the offloading method of each task; C3, C4, and C5 ensure that the UAV performs the task without exceeding its own computing capacity; C6 and C7 constrain the geographical horizontal and vertical boundaries of the UAV; and C8 ensures that each task remains within the maximum allowed delay during its execution.

[0120] Step 2: Use the two-level optimization algorithm to solve the utility function and obtain the offloading decision and resource allocation strategy of the target task.

[0121] Solving the utility function is a mixed-integer nonlinear optimization problem, and there is a significant coupling relationship between these decision-making issues. Specifically, the division of terminal areas and the planning of drone deployment directly affect the task offloading strategy and the allocation of heterogeneous computing resources. Therefore, based on this consideration, this embodiment proposes a two-level optimization algorithm, ToSMPSA, which fully considers the coupling between various decisions and divides the optimization process into two levels: upper-level optimization and lower-level optimization. The upper-level optimization uses a self-organizing map neural network model (SOM) to pre-solve the problems of terminal cluster division and drone deployment. Specifically, the number of drones and the location coordinates of the terminal devices are input into the SOM neural network model to obtain the deployment location of each drone. Based on the upper-level pre-optimization results (i.e., the drone deployment locations), the lower-level optimization uses a hybrid chaotic local search PID search algorithm (CLSPSA) to solve the utility function to determine the target task offloading decision and the resource allocation strategy for heterogeneous hardware.

[0122] Among them, the unloading decision is the unloading factor α m and uninstall method The offloading factor and offloading mode are 0 or 1, and the resource allocation strategy is

[0123] S104: Based on the offloading decision and resource allocation strategy, the target task is processed by the terminal device, drone or edge cloud server.

[0124] In the unloading factor α m When it is equal to 0, the target task is processed by the central processing unit deployed on the terminal device based on the resource allocation strategy;

[0125] In the unloading factor α m Equal to 1, uninstall method When it is equal to 1, the target task is unloaded from the terminal device to the drone, refer to Figure 3 ,Based on the heterogeneous hardware deployed on the UAV and the,resource allocation strategy, the target task is processed in stages;

[0126] In the unloading factor α m Equal to 1, uninstall method When it is equal to 0, the target task is offloaded from the terminal device to the edge cloud server through the drone, refer to Figure 3 ,Based on the heterogeneous hardware and resource allocation strategy deployed on the edge cloud server, the target task is processed in stages.

[0127] like Figure 4 As shown, it is another flow chart of the UAV-assisted edge computing method based on heterogeneous hardware provided by an embodiment of the present disclosure, and the method includes S401-S406:

[0128] S401: Establish a system model.

[0129] like Figure 2 As shown, the system model includes a three-layer architecture for collaborative task offloading and resource allocation, including the terminal device layer, the drone layer, and the cloud edge server layer.

[0130] (1) Regarding the terminal device layer:

[0131] The terminal device layer includes a large number of industrial intelligent terminals (such as robotic arms, smart cameras, AGVs) with different computing requirements, which are used to perceive and process complex data in various industrial scenarios. However, due to the limited computing power of terminal devices, computationally intensive tasks are offloaded to UAVs or edge servers for processing. The terminal devices are represented as a set D = {1, 2, ..., M}. Considering the industrial intelligent terminal device D m With limited cost, the terminal device only deploys the central processing unit (CPU) for general computing, and the parameter information is expressed as

[0132] in Indicates that each smart terminal device is equipped with CPUs with the same performance; Indicates the operating frequency of the CPU of the intelligent terminal device; and They represent the working power and idle power of the CPU of the intelligent terminal device respectively.

[0133] (2) About the drone layer:

[0134] Each drone is equipped with a MEC server with heterogeneous hardware and a communication relay function. Each drone is responsible for terminal device tasks within a specific area, with no overlapping areas. Each drone can offload terminal device tasks to the UAV for further processing or relay them to the cloud edge server base station for processing via the UAV.

[0135] There are two types of UAV communication: terminal device to UAV (G2A) and UAV to edge server base station (A2N). UAV is represented by the set U = {1,2,…,N}. UAV is equipped with a heterogeneous hardware cluster consisting of CPU, GPU, and DPU to provide services to resource-constrained mobile users. The parameters of UAV can be expressed as

[0136] in, is the number of CPUs deployed on the nth UAV, is the computing power of each CPU, and are the working power and idle power of each CPU of the UAV. The parameters of the GPU and DPU of the UAV can be expressed as The two are the same.

[0137] (3) About the cloud edge server layer:

[0138] The cloud edge server layer deploys many cloud edge servers with powerful computing capabilities at the edge of the network. When the computing resources of the drone are close to full load, the tasks can be further offloaded to the cloud edge server for processing. Similarly, the parameter information of the cloud edge server ES can be expressed as This consistency is the result of the resource sharing technology of CPU, GPU, and DPU, which leads to the discrete availability of heterogeneous hardware computing resources. Since ES has a reliable power supply infrastructure, the energy overhead of ES is not considered.

[0139] Each terminal device's tasks can be processed (executed) locally, offloaded to a UAV via a wireless channel, or further offloaded to a cloud edge server via a UAV relay for processing. This embodiment considers a 0-1 offloading approach, where each task occupies at least one CPU. A CPU, GPU, or DPU can only be assigned to one task, and the utilization of the CPU, GPU, and DPU can reach 100%.

[0140] Unloading factor α m The definition is as follows:

[0141]

[0142] Among them, 0 means that the task is processed locally by the terminal device, and 1 means that the task is offloaded to the UAV or cloud edge server ES for processing.

[0143] Uninstall method The definition is as follows:

[0144]

[0145] When α m =1, n represents the terminal device D m The area where the drone is located, Indicates that the task is offloaded to the cloud edge server ES through the UAV. Indicates that the task is offloaded to the cloud edge server UAV in the corresponding area. Similar to many other works, the transmission of the final result can be ignored because the size of the result is negligible compared to the task size and the downlink bandwidth is usually larger than the uplink bandwidth.

[0146] Drones can act as end users, offloading tasks to ground-based MEC servers for computation. Drones can also carry edge servers as aerial MEC servers, assisting ground-based terminal devices in performing computations. Drones can also act as aerial relays, forwarding tasks from ground-based terminal devices to ground-based edge servers. This establishes a three-layer collaborative edge computing network architecture: intelligent terminal layer, drone layer, and edge cloud server layer.

[0147] S402: Establish a four-stage task processing model.

[0148] like Figure 3 As shown in Figure 3, the calculation process of the target task is defined as a four-stage task processing model, including task offloading and decryption stage, task preprocessing stage, hybrid calculation stage and result aggregation stage.

[0149] Task offloading and decryption stage: The target task is offloaded and decrypted through the DPU, freeing up the CPU's computing resources, accelerating task processing, and improving overall computing efficiency.

[0150] Task preprocessing stage: Based on the requirements of the target task or the optimal use of all available heterogeneous computing resources, the CPU decomposes the decrypted target task into parallel subtasks and serial subtasks. The CPU calculates the serial subtasks and sends the parallel subtasks to the GPU for parallel computing.

[0151] Hybrid computing stage: The CPU calculates serial subtasks and obtains serial results; the GPU calculates parallel subtasks and obtains parallel results.

[0152] Result aggregation stage: The CPU further aggregates the serial results and parallel results to obtain the final result.

[0153] The target task generated by the terminal device is expressed as:

[0154] where d m Indicates the computing task amount of the terminal device, in bits; Represents the average computational effort of DPU, CPU, and GPU in cycles / bit; γ m represents the ratio of the computational load of the hybrid computing phase to the entire target task; η m It represents the ratio of the computational load of the parallel subtasks in the hybrid computing stage to the entire hybrid computing stage. By utilizing this heterogeneous computational task representation, the computational process of the target task generated by the terminal device can be mapped to a four-stage task processing model for description.

[0155] S403: Establish a communication model.

[0156] Assume that the total bandwidth is divided into two main parts: the sub-channel bandwidth B of the G2A link G , the sub-channel bandwidth B of the A2N link A BS is located at (0,0,0), UAV U n The position is represented by (x n ,y n ,h), the UAV is suspended at a certain height h. Similarly, (x m ,y m ) indicates terminal device D m The two-dimensional coordinates of .

[0157] 1) Ground-to-Air Communication: For G2A links, due to the many different objects in the real environment acting as scatterers or obstacles, the radio signals transmitted by ground-based terminal devices or drones do not propagate in free space. Instead, they may be shadowed or scattered by man-made structures, resulting in additional path loss. Therefore, the simplified free-space path loss (FSPL) model is insufficient to accurately describe the communication between ground-based terminal devices and drones. Based on this, a probabilistic path loss model that considers the probability of occurrence and path loss of LoS and Non-LoS (NLOS) communications is used to model the G2A link, providing a comprehensive description of G2A communication.

[0158] Terminal device D m and UAV U n The probability of LoS and NLoS communication between them is:

[0159]

[0160] in, a and b are constant values ​​that depend on the environment, and h is the altitude at which the UAV hovers.

[0161] In each case, the terminal device D m and UAV U n The path loss between is modeled as:

[0162]

[0163] Among them, f c is the carrier frequency, c is the speed of light. LoS and η NLoS The average path loss of the G2A link is:

[0164]

[0165] Therefore, the terminal device D m To UAV U n The data transmission rate of the uplink computing task is:

[0166]

[0167] in, Indicates terminal device D m and UAV U n The channel gain between Indicates drone U n The interference power signal of other devices in the coverage area, N0 is the noise power, p m Indicates terminal device D m The transmission power, B G Indicates the subchannel bandwidth of the G2A link.

[0168] 2) Air-to-Network Communication: For the A2N link, considering that the UAV is hovering at high altitude, the antenna reception of the cloud edge server BS is unobstructed, and the A2N communication is mainly dominated by the LOS link. Therefore, the communication environment from the UAV to the cloud edge server can be approximated as free space, and the FSPL model can be applied to A2N communication.

[0169] Therefore, UAV U n The path loss to the cloud edge server BS is:

[0170]

[0171] Similarly, the data transmission rate of the uplink computing task from the drone to the cloud edge server is:

[0172]

[0173] in, Indicates drone U n The channel gain between the cloud edge server BS, p n Indicates drone U n The transmission power, B A Indicates the sub-channel bandwidth of the A2N link.

[0174] S404: Calculate the total processing delay and total energy consumption.

[0175] First, when the unloading factor α m = 0, the local terminal device calculates the target task l m Since the terminal device is only equipped with a CPU for general computing, the task processing delay (i.e., the first delay) and energy consumption (i.e., the first energy consumption) can be expressed as:

[0176]

[0177] in represents the first delay, Indicates the average computing power of the CPU deployed on the terminal device. Indicates the CPU computing resources allocated by the terminal device to the target task, d m Indicates the computing task amount of the terminal device, Indicates the operating frequency of the CPU of the terminal device; represents the first energy consumption, Indicates the working power of the terminal device CPU. Indicates the idle power of the terminal device's CPU.

[0178] Then, when α m =1, uninstall mode When the target task is l m Unload to the UAV in the area of ​​the terminal device, and the task is unloaded to the UAV n The transmission delay on is:

[0179]

[0180] in represents the transmission delay, γ G2A (m,n) represents the terminal device D m and UAV U n The data transmission rate between m Indicates the computing task volume of the terminal device.

[0181] The task decryption phase is performed on the DPU, and the target task l m The decryption delay required for decryption is:

[0182]

[0183] in Indicates the decryption delay, represents the average computing power of the DPU deployed on the UAV, Indicates drone U n Assigned to task l m DPU computing resources, d m Indicates the computing task amount of the terminal device, Indicates the operating frequency of the DPU on the drone.

[0184] The task preprocessing stage and result aggregation stage are both performed on the CPU. The sum of the latency of the two stages is:

[0185]

[0186] in represents the two-stage delay, Indicates the average computing power of the CPU deployed on the drone, Indicates drone U n Assigned to task l m CPU computing resources, d m Indicates the computing task amount of the terminal device, Indicates the operating frequency of the CPU on the drone, γ m Indicates the ratio of the computational load of the hybrid computing phase to the entire task.

[0187] The delay of serial subtasks and parallel subtasks in the hybrid computing stage are expressed as:

[0188]

[0189] in, represents the computational delay of the serial subtask, represents the average computing power of the CPU deployed on the drone, d m Indicates the computing task amount of the terminal device, Indicates the operating frequency of the CPU on the drone, γ m Indicates the ratio of the computational load of the hybrid computing phase to the entire task, Indicates drone U n Assigned to task l m CPU computing resources, η m Indicates the proportion of the computational load of parallel subtasks in the hybrid computation.

[0190] in, represents the computational delay of parallel subtasks, represents the average computational load of the GPU deployed on the UAV, d m Indicates the computing task amount of the terminal device, Indicates the operating frequency of the GPU on the drone, γ m Indicates the ratio of the computational load of the hybrid computing phase to the entire task, Indicates drone U n Assigned to task l m GPU computing resources, η m Indicates the proportion of the computational load of parallel subtasks in the hybrid computation.

[0191] Hybrid computing latency required for hybrid computing for:

[0192]

[0193] Therefore, the task l m The total delay (second delay) and total energy consumption (second energy consumption) of offloading to the UAV for processing are:

[0194]

[0195]

[0196] in, represents the second delay, Indicates the second energy consumption.

[0197] Finally, when the unloading factor α m =1, uninstall mode When the UAV is used as an aerial relay, it will m Computation is offloaded from terminal devices to cloud edge servers.

[0198] Task 1 m The transmission delay from the terminal device to the edge server is:

[0199]

[0200] in, Represents task l m The transmission delay from the terminal device to the edge server, γ G2A (m,n) represents the terminal device D m and UAV U n The data transmission rate between A2N (n,0) represents the UAV U n Data transmission rate between the cloud edge server and the cloud edge server, d m Indicates the computing task volume of the terminal device.

[0201] Similar to drones, when tasks are offloaded to edge servers, the latency of the task decryption phase is The two-stage delay of the task preprocessing stage and the result aggregation stage is Latency of the hybrid computing stage Respectively expressed as:

[0202]

[0203] Therefore, the total delay (i.e., the third delay) and total energy consumption (i.e., the third energy consumption) required for the cloud edge server to process the target task are:

[0204]

[0205]

[0206] in, represents the third delay, Indicates the third energy consumption.

[0207] Therefore, considering the system model of the three-tier architecture, the target task is m The total processing delay and total energy consumption are:

[0208]

[0209] Among them, t m Represents the target task l m The total processing delay, α m represents the unloading factor, Indicates the uninstallation method, e m Represents the target task l m total energy consumption.

[0210] S405: Define a utility function.

[0211] In practical applications, the total processing delay and total energy consumption can have different weight coefficients. For example, when AGV is included in path planning, the system improves the delay weight to focus on delay performance. Assume that the weight coefficients of total processing delay and total energy consumption are expressed as ω t and ω e The impact of total processing delay and total energy consumption on the performance of industrial intelligent terminal equipment can be t and ω e This design can expand the model's applicability. While ensuring that all industrial intelligent terminal devices consume enough energy to complete task offload, this embodiment minimizes the total processing latency of tasks and achieves the optimal drone deployment decision (i.e., drone deployment location), resource allocation strategy, and task offloading decision.

[0212] The utility function is defined as follows:

[0213]

[0214] Among them, ω t The weight coefficient representing the impact of total processing delay on terminal device performance, ω e The weight coefficient representing the impact of total energy consumption on the performance of terminal equipment, t m represents the total processing delay, e m represents the total energy consumption, α m represents the unloading factor, Indicates the uninstall method. Represent the resource allocation strategies of the central processing unit, data processing unit and graphics processing unit respectively, {x n ,y n} indicates the deployment location of the drone.

[0215] Under the constraints of UAV computing resources, the utility function describes the UAV deployment decision-making and task computation offloading decision-making problems. UAV deployment decision-making includes determining the division of the UAV area and the geographical location of each UAV. Task computation offloading decision-making includes offloading decision-making and resource allocation strategy, that is, determining the offloading decision for each task and the computing resources allocated to the heterogeneous hardware for that task.

[0216] The utility function involves the optimization of six variables, namely α, β, k C ,k D ,k G ,{x n ,y n}, C1 and C2 determine the offloading method of each task; C3, C4, and C5 ensure that the UAV performs the task without exceeding its own computing capacity; C6 and C7 constrain the geographical horizontal and vertical boundaries of the UAV; and C8 ensures that each task remains within the maximum allowed delay during its execution.

[0217] S406: Solve the utility function to obtain the offloading decision and resource allocation strategy of the target task.

[0218] The entire model needs to address issues such as terminal area division, drone deployment, task offloading, and heterogeneous computing resource allocation decisions. Notably, this problem is a mixed-integer nonlinear optimization problem, and these decision-making issues are significantly coupled. Specifically, the terminal area division and drone deployment planning directly impact the task offloading strategy and the allocation of heterogeneous computing resources. Based on this consideration, this embodiment proposes a two-layer optimization algorithm, ToSMPSA. This algorithm fully considers the coupling between various decisions and divides the optimization process into two levels: upper-layer optimization and lower-layer optimization. The upper layer uses the SOM algorithm to solve terminal cluster division and drone deployment decisions; the lower layer uses the CLSPSA algorithm to solve task offloading and heterogeneous hardware computing resource allocation decisions. Compared with existing optimization algorithms, the ToSMPSA algorithm significantly improves solution accuracy and convergence speed, achieving optimal solutions more quickly. This allows for better support for complex scenarios involving drone collaboration and heterogeneous resource allocation, effectively increasing the computational and energy efficiency of edge computing networks and significantly improving overall system performance.

[0219] (1) About the overall framework of the two-layer optimization algorithm ToSMPSA:

[0220] like Figure 5 As shown in the figure, it is a framework diagram of the two-layer optimization algorithm, including upper-layer optimization and lower-layer optimization:

[0221] Upper layer optimization: Self-organizing map neural network (SOM) is used to divide the terminal area and make drone deployment decisions. That is, the number of drones and the location coordinates of the terminal devices are input into the self-organizing map neural network model to obtain the deployment location of each drone, such as Figure 6 The figure shows the area division and drone deployment location map.

[0222] Lower-level optimization: Given the terminal area division and UAV deployment decision (based on the deployment location of each UAV), a hybrid chaotic local search PID search algorithm (CLSPSA) is used to solve the task offloading decision and resource allocation strategy.

[0223] (2) Upper-level optimization algorithm.

[0224] The deployment decision for multiple drones is a non-convex problem. The location choices of each drone are coupled and closely related to the distribution of terminals, making it difficult to solve using conventional optimization methods. Therefore, the upper-level optimization algorithm uses a self-organizing map (SOM) neural network. SOM is a neural network-based clustering algorithm used to map high-dimensional data into a low-dimensional space while preserving the data's topological structure.

[0225] By inputting the coordinates of the terminal device and the number of drones, the upper-level optimization algorithm uses SOM to perform terminal area division and drone deployment decisions, obtains the deployment location of the drone, and then enters the lower-level optimization to obtain unloading decisions and resource allocation decisions.

[0226] The upper-level optimization algorithm includes the following steps 1-4:

[0227] Step 1: Initialize neuron weights

[0228] SOM consists of a fixed number of neurons, each of which corresponds to a weight vector w i , whose dimension is the same as the input data. Randomly initialize the weight vector w i =[ω i1 ,ω i2 ,…,ω id ], where d is the dimension of the input terminal coordinate data.

[0229] Step 2: Competition Phase

[0230] In each iteration, the input terminal coordinate data x is randomly selected, and the Euclidean distance between the input data and all neuron weight vectors is calculated to find the neuron with the smallest distance, which is called the "winning node" (Best Matching Unit, BMU).

[0231]

[0232] That is, find the i Smallest neuron c (BMU).

[0233] Step 3: Update the weights of the winner and neighbor neurons

[0234] Update the weight vectors of the BMU and its neighboring nodes to make them closer to the current input data point. At the same time, the weights of the neighboring neurons near the winner are also updated so that they are affected and gradually form a topological structure. The formula for updating the weight vector is as follows

[0235]

[0236] In the formula represents the learning rate, which gradually decreases with time t. Where η0 is the initial learning rate and τ is the time constant. ci (t) is the neighborhood function, which defines the proximity relationship between BMU node c and other nodes i. Usually, a Gaussian function is used:

[0237]

[0238] where r c and r iis the position vector of BMU node and node i. Represents the neighborhood width, which decreases gradually over time.

[0239] Step 4: Repeat

[0240] The competition and update process is repeated until the preset number of iterations is reached or convergence criteria are met. As the number of iterations increases, the learning rate η(t) and the neighborhood width σ(t) gradually decrease. Eventually, the SOM network stabilizes and outputs the corresponding terminal area division and drone deployment results.

[0241] (3) Lower layer optimization algorithm.

[0242] Based on the terminal area division and drone deployment results obtained by the upper-level optimization algorithm, the lower-level optimization algorithm uses a hybrid chaos local search PID search algorithm (CLSPSA) to solve task offloading decisions and heterogeneous computing resource allocation decisions. The hybrid chaos local search PID search algorithm, based on the concept of an incremental PID controller, aims to converge the entire population to the optimal state by continuously adjusting the system deviation. The PID search algorithm (PSA) combines the concepts of proportional, integral, and differential control to optimize the global search. Furthermore, the algorithm introduces chaos mapping and dynamic boundary adjustment. Chaos mapping fine-tunes the positions of the particles to better explore local areas, thereby improving local search capabilities. Dynamic boundary adjustment gradually narrows the search space, concentrating the search near the global optimal solution, thereby improving global search capabilities and convergence speed. The improved algorithm enhances both the local and global search capabilities of the original algorithm and improves the convergence speed, making it applicable to a variety of constrained optimization problems.

[0243] The lower-level optimization algorithm includes the following steps 1-6:

[0244] Step 1: Population initialization

[0245] First, define the decision variables, constraints, and objective function of the optimization problem. Assume that the number of decision variables is d, and the upper and lower bounds of the variables are u and l, respectively. The control parameters of CLSPSA include the maximum number of iterations T and the population size n. The initial population can be expressed as:

[0246] x ij =(u j -l j )·r1+l j ,i=1,2,…,n; j=1,2,…,d (31)

[0247] Among them, x ij represents the jth dimension of the i-th individual; u j and l jare the upper and lower bounds of the j-th dimension variable; r1 is a random number between 0 and 1.

[0248] Then, calculate the objective function value of each individual:

[0249] f i =F(x i ) (32)

[0250] Where F is the objective function, f i The objective function value of the i-th individual.

[0251] Step 2: Incremental PID Control

[0252] CLSPSA uses a discrete incremental PID algorithm to control the position update of the individual. The incremental PID algorithm adjusts the position of the individual by adjusting the three factors of proportional (P), integral (I) and differential (D). The control formula is as follows:

[0253] Δu(t)=K p [e(t)-e(t-1)]+K i e(t)+K d [e(t)-2e(t-1)+e(t-2)] (33)

[0254] Among them, K p , K i , K d are the adjustment coefficients of proportional, integral and differential respectively, and e(t) is the deviation of the system at the tth iteration.

[0255] For the minimization problem, the best individual x at iteration t is * (t) is the individual corresponding to the historical optimal value of the population. The deviation e at the tth iteration k (t) is defined as:

[0256] e j (t) = x * (t-1)-x(t-1) (34)

[0257] To facilitate calculation and iterative update, the deviation of the tth iteration can be expressed as follows:

[0258] e k-1 (t) = e k (t-1)+x * (t)-x * (t-1) (35)

[0259] In practical problems, proportional, integral and differential factors are adjusted according to different situations and problems. The calculation formula of PID control output value Δu(t) at the tth iteration is:

[0260] Δu(t)=K p ·r2·[e K (t)-e k-1 (t)]+K i ·r3·e k (t)+K d ·r4·[e k (t)-2e k-1 (t)+e k-2 (t)](36)

[0261] Among them, r2, r3, and r4 are random numbers between [0, 1], which are used to introduce random perturbations.

[0262] In order to prevent all solutions from being very close to the optimal solution, especially in the early iterations, PSA also introduces a condition factor called zero output to prevent the algorithm from falling into the local optimum. Zero output is defined as:

[0263] o(t)=(cos(1-t / T)+λr5·L)·e k (t) (37)

[0264] Where r5 is a random number vector between [0, 1], and λ is an adjustment coefficient, which is calculated as follows:

[0265]

[0266] L is the Levy flight function, which is used to introduce a large-scale random search and is defined as:

[0267]

[0268] where u and v are matrices of n rows and n columns of random numbers that follow a standard normal distribution, respectively; β is a factor set to 1.5.

[0269] Step 3: Population Update

[0270] The update of all individuals is related to Δu(t) and o(t). The population update formula is defined as:

[0271] x(t+1)=x(t)+η·Δu(t)+(1-η)·o(t) (40)

[0272] Where n is a regulating factor matrix, expressed as η=r6cos(t / T), where r6 is a random number matrix between 0 and 1.

[0273] Step 4: Chaotic Local Search

[0274] In the update mechanism of the main loop, a chaotic local search is added. In each iteration, the best performing particles in the current population (i.e., the top 30% of particles with the best fitness) are subjected to chaotic mapping. Chaotic mapping is used to fine-tune the positions of these particles to better explore the local area, thereby improving the local search capability. Assume that the solution of the current particle is x ij , the chaos calculation formula is as follows:

[0275]

[0276] Through the above-derived values, we can get the position update formula:

[0277] x′ ij =x ij +cx(j)·(u j -l j ) (42)

[0278] This formula introduces the chaos factor cx to the current solution x ij The randomness of the chaotic sequence cx enables particles to jump randomly in the search space, thus avoiding falling into the local optimal solution.

[0279] Step 5: Dynamic Boundary Adjustment

[0280] Dynamic boundary adjustment is added to the end of the main loop to dynamically adjust the boundaries of the search space. Its purpose is to adaptively shrink or expand the boundaries of the search space to accommodate the current optimal solution, allowing the algorithm to focus more closely on the possible global optimal region and accelerate convergence. Dynamic boundary adjustment is achieved by dynamically changing the upper and lower bounds of each dimension.

[0281] First, a random factor r is generated, which changes with the number of iterations t and is used to control the dynamic adjustment of the boundary. Its formula is:

[0282]

[0283] For each dimension j, the upper and lower bounds u j and l j According to the current optimal solution Make dynamic adjustments:

[0284]

[0285] Step 6: Convergence determination

[0286] By calculating the objective function value for each individual in the population, selecting the optimal solution, and determining whether convergence conditions are met (for example, reaching the maximum number of iterations or the population change is less than a certain threshold), the algorithm terminates and outputs the optimal solution, thus achieving optimal task offloading and heterogeneous computing resource allocation decisions. Otherwise, iterations continue.

[0287] According to another aspect of the embodiment of the present disclosure, a drone-assisted edge computing device based on heterogeneous hardware is provided, such as Figure 7 As shown, the device includes:

[0288] The first computing module 701 is configured to obtain a first delay and a first energy consumption when a target task is processed by a terminal device; and, based on the heterogeneous hardware deployed on the drone, process the target task in stages; and, based on the delay and energy consumption required for processing the target task at each stage, obtain a second delay and a second energy consumption when the target task is processed by the drone; and, based on the heterogeneous hardware deployed on the edge cloud server, process the target task in stages; and, based on the delay and energy consumption required for processing the target task at each stage, obtain a third delay and a third energy consumption when the target task is processed by the edge cloud server.

[0289] The second calculation module 702 is configured to obtain the total processing delay (t m ) expression; and, based on the first energy consumption, the second energy consumption and the third energy consumption, obtain the total energy consumption (e m )

[0290] An acquisition module 703 is configured to acquire an offloading decision and a resource allocation strategy for the target task based on the expression of the total processing delay and the expression of the total energy consumption; the offloading decision is used to indicate that the target task is processed by the terminal device, the drone, or the edge cloud server, and the resource allocation strategy is used to indicate the computing resources of the heterogeneous hardware to be allocated when processing the target task in stages;

[0291] The task execution module 704 is used to process the target task by the terminal device, drone or edge cloud server based on the offloading decision and resource allocation strategy.

[0292] The drone-assisted edge computing device based on heterogeneous hardware provided in the embodiments of the present disclosure and the drone-assisted edge computing method based on heterogeneous hardware provided in the embodiments of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0293] The present disclosure also provides a computer device to execute the above-mentioned UAV-assisted edge computing method based on heterogeneous hardware. Figure 8 It shows a schematic diagram of a computer device provided by some embodiments of the present disclosure. Figure 8 As shown, the computer device 8 includes: a processor 800, a memory 801, a bus 802 and a communication interface 803, and the processor 800, the communication interface 803 and the memory 801 are connected via the bus 802; the memory 801 stores a computer program that can be run on the processor 800, and when the processor 800 runs the computer program, it executes the heterogeneous hardware-based drone-assisted edge computing method provided by any of the aforementioned embodiments of the present disclosure.

[0294] The memory 801 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The communication connection between the device network element and at least one other network element is achieved through at least one communication interface 803 (which may be wired or wireless), and may use the Internet, a wide area network, a local area network, a metropolitan area network, etc.

[0295] Bus 802 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, and the like. Memory 801 is used to store programs, and processor 800 executes the programs upon receiving execution instructions. The heterogeneous hardware-based drone-assisted edge computing method disclosed in any of the aforementioned embodiments of the present disclosure may be applied to or implemented by processor 800.

[0296] The processor 800 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits or software instructions in the processor 800. The processor 800 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPTA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It may implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present disclosure may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 801 , and the processor 800 reads the information in the memory 801 and completes the steps of the above method in combination with its hardware.

[0297] The computer device provided by the embodiment of the present disclosure and the drone-assisted edge computing method based on heterogeneous hardware provided by the embodiment of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0298] An embodiment of the present disclosure also provides a computer-readable storage medium corresponding to the UAV-assisted edge computing method based on heterogeneous hardware provided in the aforementioned embodiment. The computer-readable storage medium is a CD on which a computer program (i.e., a computer program product) is stored. When the computer program is run by a processor, it will execute the UAV-assisted edge computing method based on heterogeneous hardware provided in any of the aforementioned embodiments.

[0299] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0300] The computer-readable storage medium provided by the above-mentioned embodiments of the present disclosure and the heterogeneous hardware-based drone-assisted edge computing method provided by the embodiments of the present disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0301] The present disclosure also provides a computer program product. Figure 9 The computer program product 90 carries a program code, namely a computer program 901. The instructions included in the computer program 901 can be used to execute the steps of the heterogeneous hardware-based drone-assisted edge computing method described in the above method embodiment. For details, please refer to the above method embodiment and will not be repeated here.

[0302] The computer program product may be implemented in hardware, software, or a combination thereof. In one embodiment, the computer program product is implemented as a computer storage medium. In another embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0303] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this disclosure are merely illustrative and not restrictive, and should not be construed as necessarily possessed by each embodiment of the present disclosure. Furthermore, the specific details disclosed above are provided for illustrative purposes and to facilitate understanding, rather than as limitations. These details do not limit the present disclosure to necessarily being implemented using these specific details.

[0304] The block diagrams of the devices, devices, equipment, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0305] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0306] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0307] Various changes, substitutions, and modifications may be made to the technology described herein without departing from the teachings defined by the appended claims. Moreover, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of things, means, methods, and actions described above. Currently existing or later developed processes, machines, manufactures, compositions of things, means, methods, or actions that perform substantially the same function or achieve substantially the same results as the corresponding aspects described herein may be utilized. Accordingly, the appended claims include within their scope such processes, machines, manufactures, compositions of things, means, methods, or actions.

[0308] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0309] The above description has been provided for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A UAV-assisted edge computing method based on heterogeneous hardware, characterized in that: include: Obtaining a first time delay and a first energy consumption when the terminal device processes the target task; and, based on the heterogeneous hardware deployed on the UAV, the target task is processed in stages; Based on the time delay and energy consumption required for processing the target task at each stage, obtaining a second time delay and a second energy consumption when the drone processes the target task; and,processing the target tasks in stages based on the heterogeneous hardware deployed on the edge cloud servers; Based on the delay and energy consumption required for processing the target task at each stage, obtaining a third delay and a third energy consumption when the target task is processed by the edge cloud server; Based on the first delay, the second delay, and the third delay, an expression for the total processing delay of the target task is obtained; and based on the first energy consumption, the second energy consumption, and the third energy consumption, an expression for the total energy consumption of the target task is obtained; Based on the expression of the total processing delay and the expression of the total energy consumption, obtaining an offloading decision and a resource allocation strategy for the target task; The offloading decision is used to instruct the terminal device, drone or edge cloud server to process the target task, and the resource allocation strategy is used to indicate the computing resources of heterogeneous hardware allocated when processing the target task in stages; Based on the offloading decision and resource allocation strategy, the target task is processed by the terminal device, drone or edge cloud server.

2. The method according to claim 1, wherein The heterogeneous hardware includes a central processing unit, a graphics processing unit, and a data processing unit; Based on the heterogeneous hardware deployed on the drone, the target task is processed in stages, including: Based on a data processing unit deployed on the UAV, the target task is offloaded to the UAV and the target task is decrypted; Based on a central processing unit deployed on the UAV, the decrypted target task is decomposed into parallel subtasks and serial subtasks; Performing hybrid computing by the central processing unit and the graphics processing unit, wherein the central processing unit computes serial subtasks and the graphics processing unit computes parallel subtasks; The calculation results of the serial subtasks and the parallel subtasks are aggregated by the central processing unit to obtain the average calculation amount of the central processing unit, the graphics processing unit and the data processing unit respectively.

3. The method according to claim 1, wherein Before obtaining the first latency and the first energy consumption when the terminal device processes the target task, the method further includes: A system model of drone-assisted edge computing is constructed, which includes a terminal device, a drone and an edge cloud server; wherein, a central processing unit is deployed on the terminal device; heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the drone; and heterogeneous hardware including a central processing unit, a graphics processing unit and a data processing unit is deployed on the edge cloud server.

4. The method according to claim 1, wherein Obtaining a first latency and a first energy consumption when a terminal device processes a target task, including: Obtaining a first latency based on computing resources of a central processing unit of the terminal device allocated to the target task and an average computing load of the central processing units deployed on the terminal device; A first energy consumption is obtained based on the first time delay.

5. The method according to claim 2, wherein Based on the time delay and energy consumption required for processing the target task at each stage, obtaining a second time delay and a second energy consumption when the drone processes the target task, including: Calculate the transmission delay for offloading the target task to the UAV based on the data transmission rate of the uplink computing task from the terminal device to the UAV; Calculating a decryption delay required to decrypt the target task based on computing resources of a data processing unit allocated by the drone to the target task and an average computing workload of the data processing unit; Calculating a two-stage delay required for decomposing and aggregating the target task based on computing resources of a central processing unit of the drone allocated to the target task and an average computing load of the central processing unit; Calculating the computational latency of the serial subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the central processing unit of the drone allocated to the target task, and the average computational load of the central processing unit; and calculating the computational latency of the parallel subtasks based on the proportion of the computational load of the parallel subtasks in the hybrid computation, the computational resources of the graphics processing unit of the drone allocated to the target task, and the average computational load of the graphics processing unit; Obtaining a hybrid computing delay required for hybrid computing based on the computing delay of the serial subtask and the computing delay of the parallel subtask; Based on the transmission delay, decryption delay, two-stage delay and hybrid computing delay, a second delay and a second energy consumption when the drone processes the target task are obtained.

6. The method according to claim 1, wherein Based on the expression of the total processing delay and the expression of the total energy consumption, obtaining the offloading decision and resource allocation strategy of the target task includes: Based on the expression of the total processing delay and the expression of the total energy consumption, adjusting the weight coefficients of the total processing delay and the total energy consumption, minimizing the total processing delay and the total energy consumption, and obtaining a utility function of the weighted sum of the total processing delay and the total energy consumption; The utility function is solved using a two-level optimization algorithm to obtain the offloading decision and resource allocation strategy of the target task.

7. The method according to claim 6, wherein Solving the utility function using a two-level optimization algorithm to obtain the offloading decision and resource allocation strategy for the target task includes: Inputting the number of drones and the location coordinates of the terminal devices into the self-organizing map neural network model to obtain the deployment location of each drone; Based on the deployment position of each UAV, a hybrid chaos local search PID search algorithm is used to solve the utility function to obtain the offloading decision and resource allocation strategy of the target task.

8. The method according to claim 1, wherein The uninstallation decision includes an uninstallation factor and an uninstallation mode, wherein the values ​​of the uninstallation factor and the uninstallation mode are 0 or 1; Based on the offloading decision and resource allocation strategy, the terminal device, the drone, or the edge cloud server processes the target task, including: When the offloading factor is equal to 0, based on the resource allocation strategy, the central processing unit deployed on the terminal device processes the target task; When the offloading factor is equal to 1 and the offloading mode is equal to 1, the target task is offloaded from the terminal device to the UAV, and the target task is processed in stages based on the heterogeneous hardware deployed on the UAV and the resource allocation strategy; When the offloading factor is equal to 1 and the offloading mode is equal to 0, the target task is offloaded from the terminal device to the edge cloud server through the drone, and the target task is processed in stages based on the heterogeneous hardware deployed on the edge cloud server and the resource allocation strategy.

9. A UAV-assisted edge computing device based on heterogeneous hardware, characterized in that: include: A first calculation module is used to obtain a first delay and a first energy consumption when the terminal device processes the target task; and, based on the heterogeneous hardware deployed on the UAV, the target task is processed in stages; Based on the time delay and energy consumption required for processing the target task at each stage, obtaining a second time delay and a second energy consumption when the drone processes the target task; and,processing the target tasks in stages based on the heterogeneous hardware deployed on the edge cloud servers; Based on the delay and energy consumption required for processing the target task at each stage, obtaining a third delay and a third energy consumption when the target task is processed by the edge cloud server; a second calculation module, configured to obtain an expression for a total processing delay of the target task based on the first delay, the second delay, and the third delay; and, obtaining an expression for the total energy consumption of the target task based on the first energy consumption, the second energy consumption, and the third energy consumption; An acquisition module, configured to acquire an offloading decision and a resource allocation strategy for the target task based on an expression of the total processing delay and an expression of the total energy consumption; The offloading decision is used to instruct the terminal device, drone or edge cloud server to process the target task, and the resource allocation strategy is used to indicate the computing resources of heterogeneous hardware allocated when processing the target task in stages; A task execution module is used to process the target task by the terminal device, drone or edge cloud server based on the offloading decision and resource allocation strategy.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images