Cloud edge end collaborative task offloading method and device for geographic computing multi-parallel scene

Abstract

The application discloses a cloud-edge-terminal collaborative task offloading method and device for a geographic computing multi-parallel scene and belongs to the technical field of geographic computing; the cloud-edge-terminal collaborative task offloading method for the geographic computing multi-parallel scene considers the geographic space dependence of a to-be-processed geographic task, models based on the execution mode of the geographic task, and then accurately calculates the delay and energy consumption of the geographic task in a cloud-edge-terminal architecture, finally obtains an optimal offloading decision scheme in which the energy efficiency of the geographic task in all terminals is optimal, and can effectively improve the resource utilization efficiency of the offloading decision, and solves the problem that the current cloud-edge-terminal collaborative task offloading technology adopts a coarse-grained division strategy in task division and has low offloading decision efficiency.

Description

Technical Field

[0001] This invention relates to a cloud-edge-device collaborative task offloading method and apparatus for multi-parallel geographic computing scenarios, belonging to the field of geographic computing technology. Background Technology

[0002] As the complexity and volume of geospatial computing tasks continue to grow, single computing nodes can no longer meet the real-time and efficiency requirements of task processing. Cloud-edge-device collaborative architectures combine the massive resources of cloud computing, the low latency of edge computing, and the distributed computing power of terminal devices to provide an efficient solution for large-scale geospatial computing. However, current cloud-edge-device collaborative task offloading technologies employ a coarse-grained "data proportion partitioning" strategy in task allocation, distributing the total data to the cloud, edge, and device sides according to an optimized ratio. This coarse-grained partitioning method not only ignores the differences in data transmission volume caused by various dependencies in geospatial space, such as local, neighborhood, regional, and global relationships, but also weakens the impact of geospatial computing heterogeneity on actual computing time. Furthermore, insufficient accuracy in assessing transmission and computing latency leads to a decline in the efficiency of offloading decisions. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a cloud-edge-device collaborative task offloading method and device for multi-parallel scenarios of geocomputing. It considers the geospatial dependence of the geotask to be processed, and models it based on the execution mode of the geotask. This solves the problem that the current cloud-edge-device collaborative task offloading technology adopts a coarse-grained partitioning strategy in task division, resulting in low offloading decision efficiency.

[0004] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution: This invention provides a method for offloading cloud-edge-device collaborative tasks in a multi-parallel geographic computing scenario, comprising: The execution mode of the geographic task is determined based on its geospatial dependency. Construct a latency and energy consumption analysis model for the execution mode, and calculate the latency and energy consumption of geographic tasks; For cloud-edge-device architecture, based on the latency and energy consumption of geographical tasks, a task computing resource allocation and terminal transmission power control model is established. Solve the task computational resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for geographic tasks with the best energy efficiency across all terminals; Based on the optimal offloading decision scheme, the computing frequency and transmission power allocated to the geographic task across the cloud, edge, and terminal are calculated.

[0005] Furthermore, the latency and energy consumption analysis model for constructing the execution mode, and calculating the latency and energy consumption of the geographic task, includes: Determine the task unloading strategy based on the execution mode of the geographical task; The execution modes include data parallel mode and task parallel mode; Obtain the computational prediction intensity and computational dependency characteristics of geographic tasks; Based on the computational prediction intensity and computational dependency characteristics of geographic tasks, a latency and energy consumption analysis model oriented towards their execution mode is constructed to calculate the latency and energy consumption of geographic tasks.

[0006] Furthermore, the latency and energy consumption of the computational geographic task include: The latency of a geographic task is calculated using the following formula: ; Among them, Indicates geographic tasks on the terminal With edge nodes Total delay between; This indicates the execution latency of geographic tasks on the terminal side. This indicates the execution latency of the geographic task at the edge node. This indicates the execution latency of geographic tasks on the cloud side; ; ; ; in, This indicates the computation time for the geographic task on the terminal side. This indicates the computation time of the geographic task at the edge node. This indicates the transmission time of geographic tasks between the terminal and edge nodes. This indicates the computation time for the geographic task on the cloud side. This indicates the transmission time of geographic tasks between the terminal and the cloud; The energy consumption of a geographic task is calculated using the following formula: ; in, Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates the energy consumption of geographic tasks on the terminal side. This represents the energy consumption of the geographic task at the edge node. This indicates the energy consumption of geographic tasks on the cloud side; ; ; ; in, This indicates the computational energy consumption of geographic tasks on the terminal side. This represents the computational energy consumption of the geographic task at the edge node side. This indicates the energy consumption of geographic tasks during transmission between the terminal and edge nodes. This indicates the computing power consumption of geographic tasks on the cloud side. This indicates the energy consumption for transmitting geographic tasks between the terminal and the cloud.

[0007] Furthermore, if the execution mode of the geographic task is data parallel mode, then the task unloading strategy is expressed as follows: edge nodes terminal The geographical task is denoted as: , express Input data size, Indicates the number of parts into which the geographical task is divided; In data parallel mode, the input data Divided into equal sizes A subset of the dataset, and the computational task on the subset of the dataset is denoted as: , Indicates the first division The size of the subset of data. express The size of the geospatial dependent domain data. Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; This indicates the number of geographic task subsets processed on the terminal side. This indicates the number of geographic task subsets processed on the edge node side. This indicates the number of geographic task subsets processed on the cloud side.

[0008] Furthermore, the cloud-edge-device architecture, based on the latency and energy consumption of geographical tasks, establishes a task computing resource allocation and terminal transmission power control model, including: If the execution mode of the geographic task is data parallel mode, then the task computing resource allocation and terminal transmission power control model is expressed as maximizing the overall utility of all terminal devices as the optimization objective: ; ; ; ; ; ; in, Indicates terminal transmission rate Indicates the terminal-side task computation frequency. This indicates the computation frequency of tasks on the edge node side. Indicates the computation frequency of tasks on the cloud side; V Represents the set of edge nodes. Indicates the number of edge nodes; U Represents a set of terminals. Indicates the number of terminals; Represents edge nodes terminal Geographical tasks Delayed utility weights Represents edge nodes terminal Geographical tasks Energy consumption utility weight; Represents edge nodes terminal Geographical tasks The computational time cost without task unloading. Indicates geographic tasks on the terminal With edge nodes Total delay between; Represents edge nodes terminal Geographical tasks Energy consumption cost without task unloading Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates constraints on the uninstallation strategy; This indicates a range constraint on the terminal's transmission power. Indicates the terminal's maximum transmission rate; This indicates a range constraint on the computing frequency on the cloud side. This represents the sum of the core frequencies on the cloud side; This indicates a constraint on the calculation frequency at the edge nodes. This represents the sum of the core frequencies on the edge node side; This indicates a constraint on the computing frequency on the terminal side. This represents the total core frequency on the terminal side.

[0009] Furthermore, if the execution mode of the geographic task is a task parallel mode, then the task unloading strategy is expressed as follows: Each terminal is associated with a base station, and the edge nodes terminal The geographical task is denoted as , express Input data size, Indicates the number of parallel subtasks for a geographic task; The first The task is recorded as , Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; , ; Subtasks Whether to process on the terminal side, This indicates that the subtask is unloaded on the terminal side. This indicates that the subtask is not unloaded on the terminal side; Subtasks Whether to process at the edge node side This indicates that the subtask is unloaded on the edge node side. This indicates that the subtask is not unloaded on the edge node side; Subtasks Whether to process on the cloud side, This indicates that the subtask is unloaded on the cloud side. This indicates that the subtask is not uninstalled on the cloud side.

[0010] Furthermore, the cloud-edge-device architecture-oriented model for allocating task computing resources and controlling terminal transmission power based on the latency and energy consumption of geographical tasks further includes: if the execution mode of the geographical task is a parallel task mode, then the task computing resource allocation and terminal transmission power control model is expressed as maximizing the overall utility of all terminal devices as the optimization objective. ; ; ; ; ; ; in, Indicates terminal transmission rate Indicates the terminal-side task computation frequency. This indicates the computation frequency of tasks on the edge node side. Indicates the computation frequency of tasks on the cloud side; V Represents the set of edge nodes. Indicates the number of edge nodes; U Represents a set of terminals. Indicates the number of terminals; Represents edge nodes terminal Geographical tasks Delayed utility weights Represents edge nodes terminal Geographical tasks Energy consumption utility weight; Represents edge nodes terminal Geographical tasks The computational time cost without task unloading. Indicates geographic tasks on the terminal With edge nodes Total delay between; Represents edge nodes terminal Geographical tasks Energy consumption cost without task unloading Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates constraints on the uninstallation strategy; This indicates a range constraint on the terminal's transmission power. Indicates the terminal's maximum transmission rate; This indicates a range constraint on the computing frequency on the cloud side. This represents the sum of the core frequencies on the cloud side; This indicates a constraint on the calculation frequency at the edge nodes. This represents the sum of the core frequencies on the edge node side; This indicates a constraint on the computing frequency on the terminal side. This represents the total core frequency on the terminal side.

[0011] Furthermore, the solution to the task computational resource allocation and terminal transmission power control model determines the optimal offloading decision scheme for the geographic task across all terminals with the best energy efficiency, including: Using PPO networks, a virtual mapping space is constructed for cloud-edge-device architecture; The virtual mapping space includes a state space, an action space, and a reward function. The state space includes computing resource utilization rates on the cloud, edge, and device sides, channel quality, task queue length, and time series indicators of recent task processing latency and energy consumption. The action space includes the offloading strategy action of the geographic task between the terminal and the edge node, the offloading strategy constraint, the terminal transmission power action, the terminal transmission power range constraint, the computing frequency action, the cloud-side computing frequency range constraint, the edge node-side computing frequency constraint, and the terminal-side computing frequency constraint. An optimization function is established to maximize the overall utility of all terminal devices. Negative rewards are applied to actions that violate constraints. The network is iteratively trained using the PPO algorithm until convergence is obtained to obtain the optimal solution. Output the optimal solution to obtain the optimal unloading decision scheme.

[0012] Furthermore, the PPO network includes an input layer, a shared feature extraction layer, an Actor network, a Critic network, and an output layer; The input layer, shared feature extraction layer, Actor network, and output layer are connected in sequence; the Critic network is connected to the shared feature extraction layer and the Actor network respectively. The Actor network includes an offload decision branch, a transmission power branch, and a frequency calculation branch. The Critic network outputs a scalar state value.

[0013] In another aspect, the present invention provides a cloud-edge-device collaborative task offloading device for multi-parallel geographic computing scenarios, used to implement the above-mentioned cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios, the device comprising: The task execution mode extraction module is used to determine the execution mode of a geographic task based on its geospatial dependency. The task latency and energy consumption calculation module is used to build latency and energy consumption analysis models for execution modes and calculate the latency and energy consumption of geographic tasks. The problem building module is designed for cloud-edge-device architectures, and establishes a model for task computing resource allocation and terminal transmission power control based on the latency and energy consumption of geographical tasks. The decision module is used to solve the task computing resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for the geographic task with the best energy efficiency across all terminals. The frequency and power calculation module is used to calculate the computing frequency and transmission power allocated to the cloud, edge, and terminal sides for geographic tasks based on the optimal offloading decision scheme.

[0014] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: 1. This invention considers the geospatial dependence of the geographic task to be processed, models the execution mode of the geographic task, and then accurately calculates the latency and energy consumption of the geographic task in the cloud-edge-device architecture. Finally, it obtains the optimal offloading decision scheme with the best energy efficiency of the geographic task across all terminals, which can effectively improve the resource utilization efficiency of the offloading decision and solve the problem of low offloading decision efficiency caused by the coarse-grained partitioning strategy adopted by the current cloud-edge-device collaborative task offloading technology.

[0015] 2. This invention uses a PPO network to construct a virtual mapping space for cloud-edge-device architecture and adapts to the constraints of two parallel modes. By iteratively training the network through the PPO algorithm, the joint optimization of task offloading, resource allocation and transmission control is achieved, which effectively reduces task processing latency and overall energy consumption, realizes dynamic load balancing on the cloud, edge and device sides, and improves resource utilization. Attached Figure Description

[0016] Figure 1 This is a flowchart of a cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the PPO network architecture provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the cloud-edge-device collaborative task offloading device for multi-parallel geographic computing scenarios provided in this embodiment of the invention. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention. Example 1

[0018] like Figure 1 As shown, this embodiment provides a cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios, including: Step 1: Determine the execution mode of the geographic task based on its geospatial dependency. In this embodiment, geospatial dependency includes local dependency, neighborhood dependency, regional dependency, and global dependency. If subsequent modeling only involves local dependency, no additional data transmission is required. If neighborhood, regional, or global dependency is involved, additional data transmission is required.

[0019] Step 2: Construct a latency and energy consumption analysis model for the execution mode, and calculate the latency and energy consumption of the geographic task; specifically: Determine the task unloading strategy based on the execution mode of the geographical task; The execution modes include data parallel mode and task parallel mode; If the execution mode of the geographic task is data parallel mode, then the task unloading strategy is expressed as follows: edge nodes terminal The geographical task is denoted as: , express Input data size, Indicates the number of parts into which the geographical task is divided; In data parallel mode, the input data Divided into equal sizes A subset of the dataset, and the computational task on the subset of the dataset is denoted as: , Indicates the first division The size of the subset of data. express The size of the geospatial dependent domain data. Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; This indicates the number of geographic task subsets processed on the terminal side. This indicates the number of geographic task subsets processed on the edge node side. This indicates the number of geographic task subsets processed on the cloud side; If the execution mode of the geographic task is parallel task mode, then the task unloading strategy is expressed as follows: Each terminal is associated with a base station, and the edge nodes terminal The geographical task is denoted as , Indicates the number of parallel subtasks for a geographic task; The first The task is recorded as , Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; , ; Subtasks Whether to process on the terminal side, This indicates that the subtask is unloaded on the terminal side. This indicates that the subtask is not unloaded on the terminal side; Subtasks Whether to process at the edge node side This indicates that the subtask is unloaded on the edge node side. This indicates that the subtask is not unloaded on the edge node side; Subtasks Whether to process on the cloud side, This indicates that the subtask is unloaded on the cloud side. This indicates that the subtask is not uninstalled on the cloud side; To obtain the computational prediction intensity and computational dependency characteristics of geographic tasks, this embodiment employs multimodal deep learning technology to extract features of edge geographic computing tasks and predict computational intensity in order to improve the accuracy of computational latency assessment. Specifically, it utilizes convolutional neural networks to extract geographic computing heterogeneity features and selects edge environment heterogeneity features based on hardware architecture. On this basis, an attention mechanism is used to fuse and unify the two types of features, constructing a computational intensity prediction model. The computational intensity prediction model is then used to obtain the computational prediction intensity of geographic tasks. and ; Based on the computational prediction intensity and computational dependency characteristics of geographic tasks, a latency and energy consumption analysis model is constructed for its execution mode. This model calculates the latency and energy consumption of geographic tasks, including: The latency of a geographic task is calculated using the following formula: ; in, Indicates geographic tasks on the terminal With edge nodes Total delay between; This indicates the execution latency of geographic tasks on the terminal side. This indicates the execution latency of the geographic task at the edge node. This indicates the execution latency of geographic tasks on the cloud side; ; ; ; in, This indicates the computation time for the geographic task on the terminal side. This indicates the computation time of the geographic task at the edge node. This indicates the transmission time of geographic tasks between the terminal and edge nodes. This indicates the computation time for the geographic task on the cloud side. This indicates the transmission time of geographic tasks between the terminal and the cloud; The energy consumption of a geographic task is calculated using the following formula: ; in, Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates the energy consumption of geographic tasks on the terminal side. This represents the energy consumption of the geographic task at the edge node. This indicates the energy consumption of geographic tasks on the cloud side; ; ; ; in, This indicates the computational energy consumption of geographic tasks on the terminal side. This represents the computational energy consumption of the geographic task at the edge node side. This indicates the energy consumption of geographic tasks during transmission between the terminal and edge nodes. This indicates the computing power consumption of geographic tasks on the cloud side. This indicates the energy consumption of geographic tasks during transmission between the terminal and the cloud; This embodiment takes the data parallel mode as an example: ; ; ; ; ; ; ; ; ; ; in, Indicates the terminal-side task computation frequency. This indicates the computation frequency of tasks on the edge node side. Indicates the computation frequency of tasks on the cloud side; Indicates the number of edge nodes; Indicates terminal The uplink transmission rate between the terminal and edge nodes is affected by the terminal. Transmission power The impact; Indicates assignment to the terminal The uplink transmission rate of the wired link, This indicates the power of data transmission from the edge node to the cloud. This indicates the chip manufacturing process parameters.

[0020] Step 3: For cloud-edge-device architecture, establish a task computing resource allocation and terminal transmission power control model based on the latency and energy consumption of geographical tasks; specifically: In data parallel mode, the task computing resource allocation and terminal transmission power control model is expressed as maximizing the overall utility of all terminal devices as the optimization objective: ; ; ; ; ; ; In parallel task mode, the optimization objective is: ; ; ; ; ; ; in, V Represents the set of edge nodes. Indicates the number of terminals. U Represents a set of terminals; Represents edge nodes terminal Geographical tasks Delayed utility weights Represents edge nodes terminal Geographical tasks Energy consumption utility weight; Represents edge nodes terminal Geographical tasks The computational time cost without task unloading; Represents edge nodes terminal Geographical tasks Energy consumption cost without task unloading; This indicates constraints on the uninstallation strategy; This indicates a range constraint on the terminal's transmission power. Indicates the terminal's maximum transmission rate; This indicates a range constraint on the computing frequency on the cloud side. This represents the sum of the core frequencies on the cloud side; This indicates a constraint on the calculation frequency at the edge nodes. This represents the sum of the core frequencies on the edge node side; This indicates a constraint on the computing frequency on the terminal side. This represents the total core frequency on the terminal side.

[0021] Step 4: Solve the task computational resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for the geographic task with the best energy efficiency across all terminals; specifically: Using PPO networks, a virtual mapping space is constructed for cloud-edge-device architecture; The virtual mapping space includes a state space, an action space, and a reward function. The state space includes the utilization rate of computing resources on the cloud, edge, and device sides, channel quality, task queue length, and time series indicators of recent task processing latency and energy consumption. The action space includes the offloading strategy actions, offloading strategy constraints, terminal transmission power actions, terminal transmission power range constraints, computing frequency actions, cloud-side computing frequency range constraints, edge node-side computing frequency constraints, and terminal-side computing frequency constraints for geographic tasks between the terminal and edge nodes. An optimization function is established to maximize the overall utility of all terminal devices. Negative rewards are applied to actions that violate constraints. The network is iteratively trained using the PPO algorithm until convergence is obtained to obtain the optimal solution. Output the optimal solution to obtain the optimal unloading decision scheme; like Figure 2 As shown, the PPO (Proximal Policy Optimization) network includes an input layer, a shared feature extraction layer, an Actor network, a Critic network, and an output layer. The input layer, shared feature extraction layer, Actor network, and output layer are connected in sequence; the Critic network is connected to the shared feature extraction layer and the Actor network respectively. The Actor network includes an offloading decision branch, a transmission power branch, and a computational frequency branch. The Critic network outputs a scalar state value. In data parallel mode, the output dimension of the unloading branch is The Softmax activation function is used to ensure probabilistic interpretability, and post-processing is used to ensure that constraints are met. In parallel task mode, the output dimension of the unloading branch is... The Gumbel-Softmax activation function is used to output a one-hot vector, ensuring that the constraints are satisfied. .

[0022] Step 5: Based on the optimal offloading decision scheme, calculate the computing frequency and transmission power allocated to the geographic task on the cloud, edge, and terminal sides.

[0023] Example 2 like Figure 3 As shown, a cloud-edge-device collaborative task offloading device for multi-parallel geographic computing scenarios is used to implement the cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios in Embodiment 1. The device includes: The task execution mode extraction module is used to determine the execution mode of a geographic task based on its geospatial dependency. The task latency and energy consumption calculation module is used to build latency and energy consumption analysis models for execution modes and calculate the latency and energy consumption of geographic tasks. The problem building module is designed for cloud-edge-device architectures, and establishes a model for task computing resource allocation and terminal transmission power control based on the latency and energy consumption of geographical tasks. The decision module is used to solve the task computing resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for the geographic task with the best energy efficiency across all terminals. The frequency and power calculation module is used to calculate the computing frequency and transmission power allocated to the cloud, edge, and terminal sides of the geographic task based on the optimal offloading decision scheme. The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0024] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0025] This application is described with reference to flowchart illustrations of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the processes. Figure 1 One or more processes or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0026] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0027] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0028] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for offloading cloud-edge-device collaborative tasks in multi-parallel geographic computing scenarios, characterized in that, include: The execution mode of the geographic task is determined based on its geospatial dependency. Construct a latency and energy consumption analysis model for the execution mode, and calculate the latency and energy consumption of geographic tasks; For cloud-edge-device architecture, based on the latency and energy consumption of geographical tasks, a task computing resource allocation and terminal transmission power control model is established. Solve the task computational resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for geographic tasks with the best energy efficiency across all terminals; Based on the optimal offloading decision scheme, the computing frequency and transmission power allocated to the geographic task across the cloud, edge, and terminal are calculated.

2. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 1, characterized in that, The latency and energy consumption analysis model for constructing the execution mode calculates the latency and energy consumption of geographic tasks, including: Determine the task unloading strategy based on the execution mode of the geographical task; The execution modes include data parallel mode and task parallel mode; Obtain the computational prediction intensity and computational dependency characteristics of geographic tasks; Based on the computational prediction intensity and computational dependency characteristics of geographic tasks, a latency and energy consumption analysis model oriented towards their execution mode is constructed to calculate the latency and energy consumption of geographic tasks.

3. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 2, characterized in that, The latency and energy consumption of the computational geographic task include: The latency of a geographic task is calculated using the following formula: ; in, Indicates geographic tasks on the terminal With edge nodes Total delay between; This indicates the execution latency of geographic tasks on the terminal side. This indicates the execution latency of the geographic task at the edge node. This indicates the execution latency of geographic tasks on the cloud side; ； ； ； in, This indicates the computation time for the geographic task on the terminal side. This indicates the computation time of the geographic task at the edge node. This indicates the transmission time of geographic tasks between the terminal and edge nodes. This indicates the computation time for the geographic task on the cloud side. This indicates the transmission time of geographic tasks between the terminal and the cloud; The energy consumption of a geographic task is calculated using the following formula: ; in, Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates the energy consumption of geographic tasks on the terminal side. This represents the energy consumption of the geographic task at the edge node. This indicates the energy consumption of geographic tasks on the cloud side; ； ； ； in, This indicates the computational energy consumption of geographic tasks on the terminal side. This represents the computational energy consumption of the geographic task at the edge node side. This indicates the energy consumption of geographic tasks during transmission between the terminal and edge nodes. This indicates the computing power consumption of geographic tasks on the cloud side. This indicates the energy consumption for transmitting geographic tasks between the terminal and the cloud.

4. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 2, characterized in that, If the execution mode of the geographic task is data parallel mode, then the task unloading strategy is expressed as follows: edge nodes terminal The geographical task is denoted as: , express Input data size, Indicates the number of parts into which the geographical task is divided; In data parallel mode, the input data Divided into equal sizes A subset of the dataset, and the computational task on the subset of the dataset is denoted as: , Indicates the first division The size of the subset of data. express The size of the geospatial dependent domain data. Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; This indicates the number of geographic task subsets processed on the terminal side. This indicates the number of geographic task subsets processed on the edge node side. This indicates the number of geographic task subsets processed on the cloud side.

5. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 4, characterized in that, The cloud-edge-device architecture, based on the latency and energy consumption of geographically defined tasks, establishes a model for task computing resource allocation and terminal transmission power control, including: If the execution mode of the geographic task is data parallel mode, then the task computing resource allocation and terminal transmission power control model is expressed as maximizing the overall utility of all terminal devices as the optimization objective: ； ； ； ； ； ； in, Indicates terminal transmission rate, Indicates the task computation frequency on the terminal side. This indicates the computation frequency of tasks on the edge node side. Indicates the computation frequency of tasks on the cloud side; V Represents the set of edge nodes. Indicates the number of edge nodes; U Represents a set of terminals. Indicates the number of terminals; Represents edge nodes terminal Geographical tasks Delayed utility weights Represents edge nodes terminal Geographical tasks Energy consumption utility weight; Represents edge nodes terminal Geographical tasks The computational time cost of not performing task unloading. Indicates geographic tasks on the terminal With edge nodes Total delay between; Represents edge nodes terminal Geographical tasks Energy consumption cost without task unloading Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates constraints on the uninstallation strategy; This indicates a range constraint on the terminal's transmission power. Indicates the terminal's maximum transmission rate; This indicates a range constraint on the computing frequency on the cloud side. This represents the sum of the core frequencies on the cloud side; This indicates a constraint on the computation frequency at the edge nodes. This represents the sum of the core frequencies on the edge node side; This indicates a constraint on the computing frequency on the terminal side. This represents the total core frequency on the terminal side.

6. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 2, characterized in that, If the execution mode of the geographic task is parallel task mode, then the task unloading strategy is expressed as follows: Each terminal is associated with a base station, and the edge nodes terminal The geographical task is denoted as , express Input data size, Indicates the number of parallel subtasks for a geographic task; The first The task is recorded as , Indicate processing The required number of CPU or GPU cycles to calculate the predicted strength; but: ; in, Indicates geographic tasks on the terminal With edge nodes Uninstallation strategies between different locations; , ; Subtasks Whether to process on the terminal side, This indicates that the subtask is unloaded on the terminal side. This indicates that the subtask is not unloaded on the terminal side; Subtasks Whether to process at the edge node side This indicates that the subtask is unloaded on the edge node side. This indicates that the subtask is not unloaded on the edge node side; Subtasks Whether to process on the cloud side, This indicates that the subtask is unloaded on the cloud side. This indicates that the subtask is not uninstalled on the cloud side.

7. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 6, characterized in that, The cloud-edge-device architecture, based on the latency and energy consumption of geographical tasks, establishes a task computing resource allocation and terminal transmission power control model. It further includes: if the execution mode of the geographical task is a task parallel mode, then the task computing resource allocation and terminal transmission power control model is expressed as maximizing the overall utility of all terminal devices as the optimization objective. ； ； ； ； ； ； in, Indicates terminal transmission rate, Indicates the task computation frequency on the terminal side. This indicates the computation frequency of tasks on the edge node side. Indicates the computation frequency of tasks on the cloud side; V Represents the set of edge nodes. Indicates the number of edge nodes; U Represents a set of terminals. Indicates the number of terminals; Represents edge nodes terminal Geographical tasks Delayed utility weights Represents edge nodes terminal Geographical tasks Energy consumption utility weight; Represents edge nodes terminal Geographical tasks The computational time cost of not performing task unloading. Indicates geographic tasks on the terminal With edge nodes Total delay between; Represents edge nodes terminal Geographical tasks Energy consumption cost without task unloading Indicates geographic tasks on the terminal With edge nodes Total energy consumption; This indicates constraints on the uninstallation strategy; This indicates a range constraint on the terminal's transmission power. Indicates the terminal's maximum transmission rate; This indicates a range constraint on the computing frequency on the cloud side. This represents the sum of the core frequencies on the cloud side; This indicates a constraint on the computation frequency at the edge nodes. This represents the sum of the core frequencies on the edge node side; This indicates a constraint on the computing frequency on the terminal side. This represents the total core frequency on the terminal side.

8. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 1, characterized in that, The solution task computes resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for the geographic task with the best energy efficiency across all terminals, including: Using PPO networks, a virtual mapping space is constructed for cloud-edge-device architecture; The virtual mapping space includes a state space, an action space, and a reward function. The state space includes computing resource utilization rates on the cloud, edge, and device sides, channel quality, task queue length, and time series indicators of recent task processing latency and energy consumption. The action space includes the offloading strategy action of the geographic task between the terminal and the edge node, the offloading strategy constraint, the terminal transmission power action, the terminal transmission power range constraint, the computing frequency action, the cloud-side computing frequency range constraint, the edge node-side computing frequency constraint, and the terminal-side computing frequency constraint. An optimization function is established to maximize the overall utility of all terminal devices. Negative rewards are applied to actions that violate constraints. The network is iteratively trained using the PPO algorithm until convergence is obtained to obtain the optimal solution. Output the optimal solution to obtain the optimal unloading decision scheme.

9. The cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios according to claim 8, characterized in that, The PPO network includes an input layer, a shared feature extraction layer, an Actor network, a Critic network, and an output layer. The input layer, shared feature extraction layer, Actor network, and output layer are connected in sequence; the Critic network is connected to the shared feature extraction layer and the Actor network respectively. The Actor network includes an offload decision branch, a transmission power branch, and a frequency calculation branch. The Critic network outputs a scalar state value.

10. A cloud-edge-device collaborative task offloading device for multi-parallel geographic computing scenarios, characterized in that, The apparatus for implementing the cloud-edge-device collaborative task offloading method for multi-parallel geographic computing scenarios as described in any one of claims 1 to 9, the apparatus comprising: The task execution mode extraction module is used to determine the execution mode of a geographic task based on its geospatial dependency. The task latency and energy consumption calculation module is used to build latency and energy consumption analysis models for execution modes and calculate the latency and energy consumption of geographic tasks. The problem building module is designed for cloud-edge-device architectures, and establishes a model for task computing resource allocation and terminal transmission power control based on the latency and energy consumption of geographical tasks. The decision module is used to solve the task computing resource allocation and terminal transmission power control model to determine the optimal offloading decision scheme for the geographic task with the best energy efficiency across all terminals. The frequency and power calculation module is used to calculate the computing frequency and transmission power allocated to the cloud, edge, and terminal sides for geographic tasks based on the optimal offloading decision scheme.

Images