Task offloading processing method, model training method and system

Abstract

Embodiments of the present application disclose a task offloading processing method, a model training method and system, which are used to implement task processing in the case of insufficient local computing resources of a device. The present scheme comprises: obtaining processing requirement information corresponding to a plurality of task nodes in a to-be-processed task flow and available resource information of each device in an edge network; performing feature encoding based on the feature dimensions of a preset state space to obtain task state features and environment features, and inputting the pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the to-be-processed task flow; offloading the to-be-offloaded task nodes to the corresponding target offloading device for processing based on the task offloading strategy; and generating a task processing result corresponding to the to-be-processed task flow based on the task node processing result fed back by the target offloading device.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of edge computing, and in particular to a task offloading processing method, a model training method and system. BACKGROUND

[0002] With the complication of user demand, the service functions provided by software applications are increasingly rich, which leads to an increase in task computing power demand. Although software applications change rapidly, they are often limited by hardware computing power resources in actual applications, resulting in limited actual application service functions.

[0003] For example, in an intelligent agricultural vehicle, the on-board computing power hardware is inconvenient to update, and as the requirements of on-board software services continue to increase, the on-board computing power hardware will become outdated after a few years of purchase, making it difficult to fully support the running of software application services. If the on-board computing power hardware is replaced, there may be compatibility problems such as interface mismatch.

[0004] How to achieve task processing in the case of insufficient local computing power resources of a device is a technical problem to be solved by the present application. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a task offloading processing method, a model training method and system to achieve task processing in the case of insufficient local computing power resources of a device.

[0006] In a first aspect, a task offloading processing method is provided, applied to a host device in an edge network, comprising:

[0007] Obtaining processing demand information corresponding to a plurality of task nodes in a to-be-processed task flow and available resource information of each device in the edge network;

[0008] Performing feature encoding on the processing demand information and the available resource information based on a feature dimension of a preset state space to obtain task state features corresponding to the plurality of task nodes and environment features corresponding to each device, respectively;

[0009] Inputting the task state features corresponding to the plurality of task nodes and the environment features corresponding to each device into a pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the to-be-processed task flow, the task offloading strategy including a correspondence between a to-be-offloaded task node and a target offloading device, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with a training target of shortening task processing time, and the training data includes sample task processing time, and sample state features and sample environment features corresponding to a sample task in the preset state space;

[0010] execute the task node to be unloaded based on the task offloading strategy to a corresponding target offloading device for processing;

[0011] generate a task processing result corresponding to the to-be-processed task flow based on a task node processing result fed back by the target offloading device.

[0012] In a second aspect, a task offloading processing method is provided, applied to a passive device in an edge network, including:

[0013] receiving a task node unloaded by a host device in an edge network based on a task offloading strategy, the task offloading strategy being obtained by inputting task state features corresponding to multiple task nodes in a to-be-processed task flow and environment features corresponding to each device in the edge network into a pre-trained task offloading processing model, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with a training target of shortening task processing time, and the training data includes sample task processing time, and sample state features and sample environment features corresponding to a preset state space of a sample task;

[0014] processing the task node based on local available resources to obtain a task node processing result;

[0015] feeding back the task node processing result to the host device.

[0016] In a third aspect, a training method of a task offloading processing model is provided, including:

[0017] obtaining a to-be-processed task flow sample and an edge network environment sample, the to-be-processed task flow sample including multiple sample task nodes, the multiple sample task nodes respectively corresponding to processing requirement information of a sample task, and the edge network environment sample including available resource information of multiple sample devices in an edge network;

[0018] performing feature encoding on the processing requirement information and the available resource information based on feature dimensions of a preset state space to obtain sample state features corresponding to the multiple sample task nodes and sample environment features corresponding to each sample device;

[0019] constructing a Markov decision process model with a graph structure corresponding to the sample environment features through a value memory graph algorithm;

[0020] training the Markov decision process model based on the sample state features corresponding to the multiple sample task nodes and sample task processing time with a training target of shortening task processing time to obtain the task offloading processing model.

[0021] In a fourth aspect, a task offloading processing system is provided, comprising a plurality of devices connected via edge network communication, wherein the plurality of devices include at least one host device and at least one passive device;

[0022] The host device is used to implement the steps of the method according to the first aspect;

[0023] The passive device is used to implement the steps of the method described in the second aspect.

[0024] In a fifth aspect, a task offloading processing apparatus is provided, which is applied to a host device in an edge network, including:

[0025] The first acquisition module obtains processing requirement information corresponding to multiple task nodes in the task flow to be processed and available resource information of each device in the edge network;

[0026] A first encoding module performs feature encoding on the processing requirement information and the available resource information based on a feature dimension of a preset state space to obtain task state features corresponding to multiple task nodes and environment features corresponding to each device;

[0027] A first input module inputs the task state characteristics corresponding to the plurality of task nodes and the environmental characteristics corresponding to the various devices into a pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the task process to be processed, wherein the task offloading strategy includes a correspondence between the task nodes to be offloaded and the target offloading devices, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the shortening of task processing time as a training goal, and the training data includes the sample task processing time and the sample state characteristics and sample environmental characteristics corresponding to the sample task in the preset state space;

[0028] A first offloading module, which offloads the task node to be offloaded to the corresponding target offloading device for execution based on the task offloading strategy;

[0029] The first generating module generates a task processing result corresponding to the task flow to be processed based on the task node processing result fed back by the target offloading device.

[0030] In a sixth aspect, a task offloading processing apparatus is provided, which is applied to a passive device in an edge network, including:

[0031] a second receiving module, receiving a task node offloaded by a host device in an edge network based on a task offloading strategy, wherein the task offloading strategy is obtained by the host device inputting task state characteristics corresponding to multiple task nodes in a task flow to be processed and environmental characteristics corresponding to each device in the edge network into a pre-trained task offloading processing model, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the training goal of shortening task processing time, and the training data includes sample task processing time and sample state characteristics and sample environmental characteristics corresponding to the sample task in a preset state space;

[0032] A second processing module processes the task node based on local available resources to obtain a task node processing result;

[0033] The second feedback module feeds back the task node processing result to the host device.

[0034] In a seventh aspect, a training device for a task offloading processing model is provided, comprising:

[0035] A third acquisition module acquires a sample of a task process to be processed and an edge network environment sample, wherein the sample of the task process to be processed includes multiple sample task nodes, each of which corresponds to processing requirement information of a sample task, and the edge network environment sample includes available resource information of multiple sample devices in the edge network;

[0036] a third encoding module, performing feature encoding on the processing requirement information and the available resource information based on a feature dimension of a preset state space, to obtain sample state features corresponding to a plurality of sample task nodes and sample environment features corresponding to each sample device;

[0037] The third construction module constructs a Markov decision process model with a graph structure corresponding to the sample environment characteristics through a value memory graph algorithm;

[0038] The third training module trains the Markov decision process model based on the sample state characteristics and sample task processing time corresponding to multiple sample task nodes, with the goal of shortening task processing time, to obtain the task offloading processing model.

[0039] In an eighth aspect, an electronic device is provided, which includes a processor, a memory, and a computer program stored in the memory and runnable on the processor, and when the computer program is executed by the processor, the steps of the method of the first aspect, the second aspect, or the third aspect are implemented.

[0040] In a ninth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores a computer program. The computer program is executed by a processor to implement the steps of the method in the first aspect, the second aspect, or the third aspect.

[0041] In a tenth aspect, a computer program product is provided, and the computer program product includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to cause a computer to perform the steps of the method in the first aspect, the second aspect, or the third aspect.

[0042] In the embodiments of the present application, the host device extracts features from the processing requirement information of the task nodes and the available resource information of the devices in the edge network by means of feature coding, thereby obtaining task state features and environment features of limited feature dimensions, and then inputs the task state features and the environment features into the pre-trained task offloading processing model to perform decision-making, so that a reasonable task offloading strategy for processing each task node in the task flow based on the available resources can be obtained. Subsequently, the offloading of the task nodes to be offloaded is performed according to the task offloading strategy, so that the offloaded task nodes can be processed by other devices in the edge network, thereby saving the local processing resources of the host device. Finally, the processing result corresponding to the task flow is integrated and generated according to the processing results fed back by each offloading device, so that the host device can generate the overall processing result of the task flow to be processed. The present scheme can generate a reasonable scheduling strategy for the available resources in the edge network through the pre-trained model in the case that the local resource computing power of the device is insufficient to execute the complete task flow to be processed, so as to realize the overall task flow processing at a low resource cost. Thus, the local processing resource consumption of the device is reduced, and at the same time, the local device obtains the overall task processing result of the task flow to be processed, which is beneficial to the execution of the task with higher computing power resource demand on the device with lower computing power resource. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings, which are included to provide a further understanding of the present application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and serve to explain the present application, but do not limit the present application. In the drawings:

[0044] Figure 1a is one of the flow diagrams of a task offloading processing method applied to a host device according to an embodiment of the present application;

[0045] Figure 1b is a flow diagram of a task offloading processing method according to an embodiment of the present application;

[0046] Figure 2a is another flow diagram of a task offloading processing method applied to a host device according to an embodiment of the present application;

[0047] Figure 2bThis is a schematic diagram of an application scenario of a task offloading processing method based on an intelligent cockpit system according to an embodiment of the present application;

[0048] Figure 2c is based on Figure 2b A flowchart of a task offloading processing method;

[0049] Figure 3 This is one of the flowcharts of a task offloading processing method applied to a passive device according to an embodiment of the present application;

[0050] Figure 4 This is a second flow chart of a task offloading processing method applied to a passive device according to an embodiment of the present application;

[0051] Figure 5 This is one of the flow charts of a training method for a task offloading processing model according to an embodiment of the present application;

[0052] Figure 6 This is a second flow chart of a training method for a task offloading processing model according to an embodiment of the present application;

[0053] Figure 7 This is a structural diagram of a task offloading processing system according to an embodiment of the present application;

[0054] Figure 8 This is a structural diagram of a task offloading processing device applied to a host device according to an embodiment of the present application;

[0055] Figure 9 This is a structural diagram of a task offloading processing device applied to a passive device according to an embodiment of the present application;

[0056] Figure 10 It is a structural diagram of a training device for a task offloading processing model according to an embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The figure numbers in this application are only used to distinguish the various steps in the scheme and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.

[0058] In the field of edge computing, if device software applications have high computing resource requirements, while the device hardware has low computing resource requirements, the hardware computing power often cannot meet the software requirements. To address this, device-to-device (D2D) application service computing offload can offload subtasks that are difficult to handle locally to other devices for execution.

[0059] As user needs become more complex, software applications offer increasingly richer service capabilities, leading to increased computing power requirements. While software applications evolve rapidly, practical applications are often limited by hardware computing resources, limiting the practical service capabilities that can be applied. For example, in intelligent agricultural vehicles, onboard computing hardware is difficult to update. As the requirements for onboard software services continue to increase, the hardware often lags behind within a few years of purchase, making it difficult to fully support the operation of software application services. Replacing the onboard computing hardware can also lead to compatibility issues such as interface mismatches.

[0060] How to achieve task processing when the local computing power resources of the device are insufficient is the technical problem to be solved by this application.

[0061] The solutions provided in the embodiments of this application relate to the fields of D2D collaborative task processing, task offloading scheduling strategy, distributed computing, and edge computing services for intelligent agricultural machinery vehicles.

[0062] The embodiment of the present application provides a task offloading processing method, which is applied to a host device in an edge network, such as Figure 1a Shown, including:

[0063] S11: Obtain processing requirement information corresponding to multiple task nodes in the task process to be processed and available resource information of each device in the edge network.

[0064] The pending task flow may be generated or received by the host device and may include multiple task nodes and the processing order of these task nodes. Each task node may carry processing requirement information for the corresponding task. The processing requirement information may include information such as the task type, processing rules, and resources required to process the task. The processing requirement information may be used to implement processing on the corresponding task node.

[0065] The available resource information of each device in the edge network may specifically include the available processor resources, memory resources, etc. This available resource information is used to indicate the device's processing capability for tasks. If there are many tasks to be processed, this available resource information may also include queue information for tasks waiting to be processed.

[0066] For example, the solution provided in the embodiments of the present application can be applied to a smart cockpit system, which includes a vehicle terminal and electronic terminals for the vehicle's drivers and passengers, such as smartphones, tablet computers, and laptops. The vehicle terminal and electronic terminals are connected via an edge network communication connection. Any device in the edge network can act as a host device to offload tasks to other devices in the same edge network, thereby meeting the computing needs of software applications through D2D collaborative processing.

[0067] If the vehicle terminal is the host device, in this step, the vehicle terminal can obtain the available resource information of the electronic terminals of the driver and passengers on the same edge network through the edge network. This available resource information can be transmitted by broadcasting within the edge network, or the host device and each passive device can establish a point-to-point communication link based on the edge network to achieve the required information transmission through the point-to-point communication link.

[0068] S12: Based on the feature dimension of the preset state space, feature encoding is performed on the processing requirement information and the available resource information to obtain task state features corresponding to multiple task nodes and environment features corresponding to each device.

[0069] The preset state space can be set according to the process of the task to be processed. For example, the characteristic dimensions in the preset state space can be used to represent the resource requirements of the task in terms of processor resources, memory resources, time resources, etc., and to represent the amount of processor resources, memory resources, etc. available for the device to process the task.

[0070] For example, the state space includes 38-dimensional features as shown in the following table:

[0071]

[0072]

[0073] In this step, feature encoding is used to transform the processing requirements of task nodes into finite-dimensional task state features within a pre-defined state space. Furthermore, the available resource information corresponding to the device is transformed into finite-dimensional environmental features within the pre-defined state space. Making decisions based on these finite-dimensional features effectively improves the efficiency of model-generated task offloading strategies in subsequent steps.

[0074] In actual application, the to-be-processed task flow of the task can be input into the Embedding encoder to obtain the task state features corresponding to each task node, and the available resource information of each device in the edge network can be input into the Embedding encoder to obtain the environment features corresponding to each device. The to-be-processed task flow can be in the form of a directed acyclic graph.

[0075] S13: input the task state features corresponding to each task node and the environment features corresponding to each device into the pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the to-be-processed task flow, the task offloading strategy including the correspondence between the to-be-offloaded task node and the target offloading device, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the training target of shortening the task processing time consumption, and the training data includes the sample task processing time consumption, and the sample state features and sample environment features corresponding to the sample task in the preset state space.

[0076] In this step, the coded features are input into the pre-trained task offloading processing model, and the task offloading processing model generates a task offloading strategy according to the task to be processed and the available resources possessed by the edge network. The task offloading strategy includes the correspondence between the to-be-offloaded task node and the target offloading device, that is, the corresponding offloading task node is processed by the available resources possessed by the target offloading device, so as to realize the processing of the to-be-offloaded task node.

[0077] In the task offloading strategy, local task nodes can also be included, and these local task nodes can be executed by the local available resources of the host device without performing task offloading.

[0078] The task offloading processing model is obtained by training a value memory graph model based on training data with the training target of shortening the task processing time consumption, wherein the training data includes the sample task processing time consumption, and the sample state features and sample environment features corresponding to the sample task in the preset state space. The training steps of the model will be described in detail later.

[0079] S14: offload the to-be-offloaded task node to the corresponding target offloading device for processing based on the task offloading strategy.

[0080] In this step, the to-be-offloaded task node is offloaded to the corresponding target offloading device in the strategy according to the task offloading strategy output by the model in the above step, so as to execute the processing of each task node in the to-be-processed task flow by each device in the edge network.

[0081] S15: generate a task processing result corresponding to the to-be-processed task flow based on the task node processing result fed back by the target offloading device.

[0082] In this step, the task node processing results fed back by each target offloading device are received, wherein the obtained task node processing results can be integrated according to the task flow to be processed, thereby generating a task processing result corresponding to the task flow to be processed.

[0083] Optionally, if the host device processes a portion of the task nodes through local resources, in this step the task processing results obtained by local processing are integrated with the task processing results fed back by the target offloading device in the above-mentioned edge network to generate the task processing results corresponding to the task flow to be processed.

[0084] Next, combine Figure 1b To further explain this plan, Figure 1b The corresponding relationship between each step mark and the function or module of the execution step is shown in the following table:

[0085]

[0086] Figure 1b The present invention shows a flow chart of a task offloading processing method, which includes a computing resource device (1) as a host device (Master) and a computing resource device (2) as a passive device (Slaver).

[0087] In this example, the computing resource device (1) is in the Master working state. It is assumed that the task flow to be processed in the computing resource device (1) includes task (1), task (2), etc.

[0088] First, the host device uses a compiler to parse and decompose the task into a directed acyclic graph (DAG) through step ①. Each node in the DAG records the processing requirement information of the subtask corresponding to the node. Specifically, it may include the processing time limit information of the task, the processor resource information required to process the task, the memory resource information required to process the task, etc.

[0089] Then, through step ②, the processing requirement information of the DAG task and the available resource information of each device in the edge network are input into the local trainer to obtain the task offloading strategy output by the local trainer decision.

[0090] Then, the offloading scheduler schedules and allocates the DAG tasks according to the task offloading strategy, wherein the DAG tasks decided to be executed locally are provided to the local executor through step ⑤, and the DAG tasks decided to be offloaded are provided to the remote transmission unit through step ⑥, thereby offloading the task nodes to be offloaded to the corresponding target offloading device for execution through the transmission method shown in ⑦. After any target offloading device completes the execution of the received task, it can feed back the task processing result to the remote transmission unit of the host device, and then through step The processing results of each offload task are provided to the local executor, and the local executor summarizes the processing results of each DAG task processed locally and offloaded, thereby completing the process through steps Execution output.

[0091] Through the solution provided by the embodiments of the present application, the host device extracts features from the processing requirement information of the task node and the available resource information of the devices in the edge network through feature encoding, thereby obtaining task state features and environmental features of limited feature dimensions, which are then input into the pre-trained task offloading processing model for execution decision-making, and a reasonable task offloading strategy can be obtained for each task node in the task process based on available resources. Subsequently, the task nodes to be offloaded are offloaded according to the task offloading strategy, and the offloaded task nodes can be processed by other devices in the edge network, saving the local processing resources of the host device. Finally, the processing results corresponding to the task process are integrated based on the processing results fed back by each offloading device, enabling the host device to generate the processing results of the entire task process to be processed. This solution can generate a reasonable scheduling strategy for the available resources in the edge network through a pre-trained model when the local resource computing power of the device is insufficient to execute the entire task process to be processed, and realize the overall task process processing at a low resource cost. This reduces the consumption of local processing resources of the device while enabling the local device to obtain the task processing results of the entire task process to be processed, which can facilitate the execution of tasks with high computing resource requirements in devices with low computing resource resources.

[0092] Based on the solution provided in the above embodiment, optionally, Figure 2a As shown, in the above step S11, the processing requirement information corresponding to the multiple task nodes in the task flow to be processed and the available resource information of each device in the edge network are obtained, including:

[0093] S21: Compiling the task flow to be processed into a directed acyclic graph through a task compiler. The directed acyclic graph includes a plurality of task nodes connected by directed lines, and any task node carries corresponding processing requirement information.

[0094] See also Figure 1bIn step 1 shown in the middle, the task flow to be processed is parsed by a compiler to obtain a directed acyclic graph (DAG) form of the task for subsequent decision processing of each node in the DAG task.

[0095] S22: Obtain available resource information of each device in the edge network through global resource broadcast of the edge network.

[0096] In the embodiments of the present application, the available resource information of each device in the edge network can be obtained through a broadcast manner. Any device in the edge network can broadcast its available resource information based on a preset period or in real time. The available resource information can include available processor resources, available memory resources, task queuing status and other information, which indicates the load capacity of the device for tasks.

[0097] The scheme provided by the embodiments of the present application can be applied to the application scenario of task offloading between user equipment (User Equipment, UE), and can be used for task offloading from a local device UE-local to other devices UE-others.

[0098] For example, the embodiments of the present application can be applied to a system based on a vehicle terminal, which can include a driving and control system, a car head unit, a notebook computer, a smart phone and a tablet computer, and the above-mentioned devices in the system are connected based on an edge network. The above-mentioned devices in the system can be provided with the required functions for task offloading by installing an application program, so as to implement the scheme provided by the embodiments of the present application.

[0099] Figure 2b An application scenario schematic diagram of an intelligent cockpit system is shown, wherein the task flow to be executed of the intelligent cockpit system can include a personnel identity recognition task and an in-vehicle state analysis task, and the above-mentioned task flow to be executed is shown in the form of a directed acyclic graph in Figure 2b , which includes a plurality of sub-task nodes in the task flow to be executed and a directed line connecting the sub-task nodes.

[0100] Based on Figure 2b the shown scenario, the flowchart of performing task offloading processing by the scheme provided by the embodiments of the present application is shown in Figure 2c . Through the scheme provided by the embodiments of the present application, the computing offloading dynamic scheduling of the task can be performed according to the real-time change of the computing power resources, so as to periodically and dynamically implement the task offloading processing.

[0101] Optionally, the processing period of the to-be-processed task flow is divided, for example, all sub-tasks in a first to-be-processed task flow are processed by the first round of task computation offloading dynamic scheduling, all sub-tasks in a second to-be-processed task flow are processed by the second round of task computation offloading dynamic scheduling, and so on.

[0102] Figure 2c In the illustrated DAG, dark nodes represent tasks executed on local devices, and light gray nodes represent tasks offloaded to other devices for execution.

[0103] In the first round of task computation offloading dynamic scheduling, the computation offloading request is initiated by the driving and control system, that is, the driving and control system is in the host state and is the host device in the edge network. The in-vehicle infotainment system, the notebook computer, the smartphone, and the tablet computer assist the host device to perform task processing as passive devices. After determining the task offloading strategy, the host device offloads the to-be-offloaded tasks to the corresponding target offloading devices and recovers the processing results of the offloaded tasks, thereby realizing distributed processing of tasks.

[0104] In the second round of task computation offloading dynamic scheduling, the computation offloading request is initiated by the in-vehicle infotainment system, and the driving and control system, the notebook computer, the smartphone, and the tablet computer assist the host device to perform task processing as passive devices.

[0105] In actual applications, the host device in the system can be the device that initiates the to-be-processed task flow, or the host device can be the device that first receives the to-be-processed task flow in the system, that is, each device in the system can be switched between the host device and the passive device. In addition, the available computing power resources of any device in the system change in real time based on various events such as task analysis, task offloading, task processing, and processing result integration, and therefore, the same to-be-processed task flow executed on the same device can result in different task offloading strategies. Through the scheme provided in the embodiments of the present application, the task offloading strategy is dynamically generated, which can make the task offloading strategy conform to the available resource state of each device in the system and improve the overall efficiency of task processing.

[0106] The embodiments of the present application provide a task offloading processing method applied to a passive device in an edge network, such as Figure 3 As shown, the method comprises the following steps:

[0107] S31: Receive a task node that is unloaded by a host device in an edge network based on a task offloading strategy, wherein the task offloading strategy is obtained by the host device inputting task state characteristics corresponding to multiple task nodes in the task process to be processed and environmental characteristics corresponding to each device in the edge network into a pre-trained task offloading processing model, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the training goal of shortening task processing time, and the training data includes sample task processing time and sample state characteristics and sample environmental characteristics corresponding to the sample task in a preset state space.

[0108] S32: Process the task node based on local available resources to obtain a task node processing result.

[0109] S33: Feedback the task node processing result to the host device.

[0110] On the passive device side, for the task nodes offloaded by the host device, processing is performed through local resources, and the processing results of the task nodes obtained are fed back to the host device that initiated the task offloading, so that the host device can generate the overall processing results of the task process to be processed when its own processing resources are insufficient, and realize the overall task process processing at a low resource cost.

[0111] Based on the solution provided in the above embodiment, optionally, before the above step S32, that is, before processing the task node based on the local available resources, as shown in FIG. Figure 4 As shown, it also includes:

[0112] S41: When there are multiple received task nodes to be processed, the congestion manager sequentially allocates the task nodes to be processed to locally available virtual machines.

[0113] In this step, the received offload tasks of the host device are arranged by the congestion manager to achieve sequential processing. The congestion manager can achieve sequential processing of tasks by controlling the queue (Ready_Task).

[0114] In the example of this application, any passive device can split local computing resources through virtual machines. In this solution, by deploying virtual machines, a powerful computing device can be split into multiple virtual sub-devices to improve the resource utilization of this powerful computing device. When executing task processing, a task to be processed is assigned to a virtual machine for processing by remotely executing service instructions. In this way, only a small part of the resources of this powerful computing device is required, and other resources (i.e., virtual machines with no assigned tasks) are idle and can be used to process subsequent tasks in the queue, thereby realizing parallel processing of multiple tasks in this powerful computing device and improving resource utilization.

[0115] Optionally, a fixed number of thread pools can be used in actual applications to simulate the role of a virtual machine.

[0116] Ready_Task can poll the tasks to be processed in sequence, see Figure 1b For the tasks to be processed that are polled, the tasks are provided to the remote execution service through step ⑧, and the remote execution service allocates the tasks to the virtual machine for execution and processing through step ⑨.

[0117] In the above step S32, feeding back the task node processing result to the host device includes:

[0118] S42: Feedback the processing result of the local available virtual machine on the task node to be processed to the host device through the congestion manager.

[0119] See also Figure 1b After the virtual machine completes the task processing, it feeds the task processing result back to the remote execution service through step ⑩, and then the remote execution service passes step The task processing result is fed back to the congestion manager, and then fed back to the remote transmission unit of the host device through the transmission method shown in ⑦.

[0120] Through the solution provided in the embodiment of the present application, the passive device can process the offloaded tasks in an orderly manner. When there are a large number of tasks to be processed, through the solution provided in the embodiment of the present application, the virtual machine can be used to reasonably call and allocate local available resources, thereby improving task throughput and overall task processing efficiency.

[0121] Task offloading can be achieved using rule-based algorithms. However, this approach has certain limitations and drawbacks. On the one hand, the main drawback of rule-based algorithms is their difficulty handling complex tasks and environments. On the other hand, because rules can only be applied within predefined situations, rule-based algorithms are also unable to adapt to unknown situations and changes. This limits the scope and adaptability of rule-based algorithms in practical applications.

[0122] Task offloading can also be achieved using heuristic algorithms, which utilize heuristic rules for decision-making, such as greedy algorithms and genetic algorithms. Compared to the aforementioned rule-based approaches, this approach can adaptively learn the characteristics of the task and environment. However, its disadvantage is that the performance of heuristic algorithms is highly dependent on the characteristics of the specific task, making it impossible to perform universal adaptive learning.

[0123] Compared to the above-mentioned rule-based and heuristic algorithm methods, reinforcement learning-based methods are more versatile and adaptable, can handle complex tasks and environments, and have good generalization capabilities. However, task offloading methods based on reinforcement learning still face some challenges. On the one hand, the task decomposition problem has high dimensionality and nonlinear characteristics, which limits the reinforcement learning algorithm in solving such problems. On the other hand, reinforcement learning-based methods need to process large-scale state spaces, which increases the computational complexity of the algorithm. In addition, reinforcement learning-based methods are also limited by environmental uncertainty, because changes in the environment will cause the strategy to be re-optimized, which will reduce the efficiency of the algorithm.

[0124] In summary, there are the following obstacles to the implementation of task offloading:

[0125] First, the task offloading method based on rule-based algorithms has difficulty in handling complex tasks and environments and cannot adapt to unknown situations and changes.

[0126] Secondly, task offloading methods based on heuristic algorithms are highly dependent on the characteristics of specific tasks and cannot perform general adaptive learning.

[0127] Third, the task offloading method based on reinforcement learning will be limited when dealing with task decomposition problems with high dimensionality and nonlinear characteristics, and the computational complexity of the algorithm will increase exponentially.

[0128] In order to solve the problems existing in the related art, an embodiment of the present application provides a training method for a task offloading processing model, which is used to implement task offloading processing decisions based on a value memory graph and nearest neighbor strategy optimization.

[0129] The present application embodiment provides a method for training a task offloading processing model, such as Figure 5 Shown, including:

[0130] S51: Obtain a sample of the task process to be processed and an edge network environment sample, wherein the sample of the task process to be processed includes multiple sample task nodes, and the multiple sample task nodes respectively correspond to processing requirement information of the sample tasks, and the edge network environment sample includes available resource information of multiple sample devices in the edge network.

[0131] In this step, data samples are obtained for model building and training. Specifically, these samples include samples of pending task processes and edge network environment samples. The pending task process samples include processing requirement information for the sample tasks, indicating the conditions required for task processing. The edge network environment samples include available resource information for the sample devices, indicating the resources available on the sample devices for task processing.

[0132] S52: performing feature coding on the processing requirement information and the available resource information based on the feature dimension of a preset state space, to obtain sample state features corresponding to a plurality of sample task nodes and sample environment features corresponding to each sample device.

[0133] In this step, the processing requirements of the DAG task can be modeled and learned through the Value Memory Graph (VMG) Embedding based on the preset state space. The processing requirements information can specifically include the processing time limit of the subtask, the processor resources and memory resources required for task processing, etc.

[0134] S53: Constructing a Markov decision process model with a graph structure corresponding to the sample environment characteristics through a value memory graph algorithm.

[0135] In this step, a VMG-based Markov decision process (MDP) is constructed based on the preset state space. VMG is a Markov decision process with a graph structure. It is a simplified version of some states in the original environment and has a discrete and relatively small state-action space.

[0136] Specifically, the partial state mentioned above may include the task processing time limit t, the amount of processor resources C required for the task, and the amount of memory M. For example, in the step of constructing a VMG-based MDP based on the state space, the VMG Embedding encoding of the task state (t, C, M) in an environment is first learned. Next, a mapping is constructed from the VMG Embedding encoding to task processing actions, where task processing actions include calling local resources to process the task or offloading the task to a passive device for execution. Finally, a Markov decision process is defined on a graph as an abstract representation of the partial state of the offloaded environment.

[0137] S54: Based on the sample state features and sample task processing times respectively corresponding to a plurality of sample task nodes, the Markov decision process model is trained with shortening the task processing time as a training goal to obtain the task offloading processing model.

[0138] In the solution provided by the embodiment of the present application, the value memory graph (VMG) is used to represent the original partial offloading environment as a graph-based MDP with relatively small and discrete action and state spaces, which effectively improves the quality of the model. In the process of performing reinforcement learning (RL), combined with the offloading environment implementation, the difficulty of strategy learning can be effectively reduced. Among them, the PPO-Clip algorithm is used to repeatedly utilize the sampled DAG to effectively improve the sample utilization rate. Compared with the reinforcement learning methods of rules, heuristics and related technologies, the solution provided by the embodiment of the present application significantly improves the time efficiency of D2D edge network task offloading.

[0139] Optional, such as Figure 6 As shown, in the above step S54, based on the sample state characteristics and sample task processing time corresponding to the plurality of sample task nodes, the Markov decision process model is trained with the training goal of shortening the task processing time to obtain the task offloading processing model, including:

[0140] S61: Based on the sample state characteristics and sample task processing time corresponding to multiple sample task nodes, the training target is adjusted through a proximal strategy optimization algorithm, and the Markov decision process model is iteratively trained to obtain the task offloading processing model.

[0141] In the solution provided by the embodiment of the present application, the task offloading processing model in the embodiment of the present application can be trained using the VMGPPO algorithm. In which, the PPO-Clip method is used to train the above VMG to obtain the task offloading processing model.

[0142] In order to optimize the rationality of the task offloading strategy output by the model, the task offloading processing model in the embodiment of the present application adopts the Value Memory Graph Proximal Policy Optimization (VMGPPO) algorithm for training.

[0143] Specifically, this scheme uses a value memory graph to represent the original partial offloading environment as a graph-based Markov decision process with relatively small and discrete action and state spaces, and uses the PPO-Clip algorithm to reuse the sampled DAG multiple times to solve the problem of low sample utilization, thereby improving the rationality of the task offloading strategy output by the trained task offloading processing model.

[0144] In practical applications, the task offloading processing model trained through the embodiment of the present application determines the task offloading strategy applicable to the task flow to be processed, and can use the dynamic scheduling strategy to optimize the waiting time of the dependent constrained subtasks in each DAG task, while minimizing the execution time of each DAG task.

[0145] The embodiment of the present application provides a task offloading processing system, such as Figure 7 As shown, it includes multiple devices connected based on edge network communication, and the multiple devices include at least one host device 71 and at least one passive device 72;

[0146] The host device 71 is used to implement the steps of the task offloading processing method applied to the host device in the edge network in the above embodiment;

[0147] The passive device 72 is used to implement the steps of the task offloading processing method applied to the passive device in the edge network in the above embodiment.

[0148] Through the system provided in the embodiment of the present application, D2D task offloading processing can be realized based on the edge network, and the host device's pending task flow can be collaboratively processed based on the available resources of each device in the edge network. While reducing the host device's processing resource consumption, the host device can obtain the overall task processing results of the pending task flow, which can be beneficial to the execution of tasks with high computing power resource requirements in host devices with low computing power resources.

[0149] Based on the system provided by the above embodiment, optionally, the host device includes a vehicle-mounted terminal, and the passive device includes a mobile terminal directly connected to the vehicle-mounted terminal for communication.

[0150] In intelligent agricultural machinery vehicles, the on-board computing power hardware is not easy to update. As the requirements for on-board software services continue to increase, the on-board computing power hardware will lag behind the requirements of software services after a few years of purchase, making it difficult to fully support the operation of software application services. Through the solution provided in the embodiment of the present application, the on-board terminal can offload the tasks it needs to perform to the mobile terminal in the edge network, thereby utilizing the resources of the mobile terminal to process the tasks and enabling the on-board terminal to obtain the overall processing results of the tasks. This solution can effectively reduce the processing resource consumption of the on-board terminal, allowing the on-board terminal to obtain the overall task processing results of the task flow to be processed, which can be beneficial to the execution of tasks with high computing power resource requirements in on-board terminals with low computing power resources.

[0151] The embodiment of the present application can be used to solve the technical problem of implementing task processing when the local computing power resources of the device are insufficient, and provides a task offloading processing device 80, such as Figure 8 As shown, the host devices used in the edge network include:

[0152] The first acquisition module 81 acquires processing requirement information corresponding to multiple task nodes in the task flow to be processed and available resource information of each device in the edge network;

[0153] A first encoding module 82 performs feature encoding on the processing requirement information and the available resource information based on a feature dimension of a preset state space to obtain task state features corresponding to multiple task nodes and environment features corresponding to each device;

[0154] The first input module 83 inputs the task state characteristics corresponding to the multiple task nodes and the environmental characteristics corresponding to the various devices into a pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the task process to be processed, wherein the task offloading strategy includes a correspondence between the task nodes to be offloaded and the target offloading devices, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the shortening of task processing time as a training goal, and the training data includes the sample task processing time and the sample state characteristics and sample environmental characteristics corresponding to the sample task in the preset state space;

[0155] A first offloading module 84 is configured to offload the task node to be offloaded to a corresponding target offloading device for execution based on the task offloading strategy;

[0156] The first generating module 85 generates a task processing result corresponding to the task flow to be processed based on the task node processing result fed back by the target offloading device.

[0157] Through the solution provided by the embodiments of the present application, the host device extracts features from the processing requirement information of the task node and the available resource information of the devices in the edge network through feature encoding, thereby obtaining task state features and environmental features of limited feature dimensions, which are then input into the pre-trained task offloading processing model for execution decision-making, and a reasonable task offloading strategy can be obtained for each task node in the task process based on available resources. Subsequently, the task nodes to be offloaded are offloaded according to the task offloading strategy, and the offloaded task nodes can be processed by other devices in the edge network, saving the local processing resources of the host device. Finally, the processing results corresponding to the task process are integrated based on the processing results fed back by each offloading device, enabling the host device to generate the processing results of the entire task process to be processed. This solution can generate a reasonable scheduling strategy for the available resources in the edge network through a pre-trained model when the local resource computing power of the device is insufficient to execute the entire task process to be processed, and realize the overall task process processing at a low resource cost. This reduces the consumption of local processing resources of the device while enabling the local device to obtain the task processing results of the entire task process to be processed, which can facilitate the execution of tasks with high computing resource requirements in devices with low computing resource resources.

[0158] The modules in the device provided in the embodiments of the present application can also implement the method steps provided in the method for task offloading processing of the host device in the edge network. Alternatively, the device provided in the embodiments of the present application can further include other modules in addition to the above modules to implement the method steps provided in the method for task offloading processing of the host device in the edge network. The device provided in the embodiments of the present application can achieve the technical effects achieved by the method for task offloading processing of the host device in the edge network.

[0159] The embodiments of the present application can be used to solve the technical problem of implementing task processing in the case of insufficient local computing power resources of the device, and provide a task offloading processing device 90, as shown in Figure 9 The device for passive equipment in the edge network includes:

[0160] The second receiving module 91 receives the task node offloaded by the host device in the edge network based on the task offloading strategy, wherein the task offloading strategy is obtained by inputting the task state features corresponding to a plurality of task nodes in a task processing flow and the environment features corresponding to each device in the edge network into a pre-trained task offloading processing model, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the training target of shortening task processing time, and the training data includes sample task processing time, sample state features and sample environment features corresponding to the sample task in a preset state space;

[0161] The second processing module 92 processes the task node based on the local available resources to obtain a task node processing result.

[0162] The second feedback module 93 feeds back the task node processing result to the host device.

[0163] By means provided by the embodiments of the present application, the passive device extracts features from the processing requirement information of the task node and the available resource information of the device in the edge network by means of feature coding, thereby obtaining task state features and environment features of limited feature dimensions, and then inputting the features into the pre-trained task offloading processing model to perform decision-making, so that a reasonable task offloading strategy for each task node in the processing task flow based on the available resources can be obtained. Subsequently, the offloading is performed on the task node to be offloaded according to the task offloading strategy, so that the offloaded task node can be processed by other devices in the edge network, thereby saving the local processing resources of the host device. Finally, the processing result corresponding to the task flow is integrated and generated according to the processing results fed back by each offloading device, so that the host device can generate the overall processing result of the task flow to be processed. The present scheme can generate a reasonable scheduling strategy for the available resources in the edge network through the pre-trained model in the case that the local resource computing power of the device is insufficient to execute the complete task flow to be processed, so as to realize the overall task flow processing at a low resource cost. Thus, the local processing resource consumption of the device is reduced, and at the same time, the local device obtains the overall task processing result of the task flow to be processed, which is beneficial to the execution of the task with higher computing power resource demand on the device with lower computing power resource.

[0164] With the complication of user demand, the service functions provided by software applications are increasingly rich, which leads to an increase in task computing power demand. Although software applications change rapidly, they are often limited by hardware computing power resources in actual application, resulting in limited actual applicable service functions.

[0165] For example, in an intelligent agricultural vehicle, the on-board computing power hardware is inconvenient to update, and as the requirements of on-board software services increase, the on-board computing power hardware will become outdated after being purchased for a few years, which makes it difficult to fully support the running of software application services. If the on-board computing power hardware is replaced, there may be compatibility problems such as interface mismatch.

[0166] The embodiments of the present application can be used to solve the technical problem of implementing task processing in the case of insufficient local computing power resources of the device, and provide a training device 100 of a task offloading processing model, as shown in Figure 10 The training device 100 comprises:

[0167] A third acquisition module 101 acquires a task flow to be processed sample and an edge network environment sample, the task flow to be processed sample comprising a plurality of sample task nodes, the plurality of sample task nodes respectively corresponding to processing requirement information of a sample task, and the edge network environment sample comprising available resource information of a plurality of sample devices in the edge network;

[0168] The third encoding module 102 performs feature encoding on the processing requirement information and the available resource information based on the feature dimension of the preset state space, thereby obtaining sample state features corresponding to a plurality of sample task nodes and sample environment features corresponding to each sample device.

[0169] The third construction module 103 constructs a Markov decision process model with a graph structure corresponding to the sample environment characteristics through a value memory graph algorithm;

[0170] The third training module 104 trains the Markov decision process model based on the sample state features and sample task processing times corresponding to a plurality of sample task nodes, with the goal of shortening task processing time, to obtain the task offloading processing model.

[0171] Through the solution provided by the embodiment of the present application, the original partial offloading environment is represented as a graph-based MDP with a relatively small and discrete action and state space using a value memory graph, which effectively improves the quality of the model. During the model training process, the PPO-Clip algorithm is used to repeatedly utilize the sampled DAG to effectively improve the sample utilization rate. Compared with the reinforcement learning methods of rule-based, heuristic and related technologies, the solution provided by the embodiment of the present application significantly improves the time efficiency of D2D edge network task offloading. Among them, the use of the PPO-Clip algorithm to repeatedly utilize the sampled DAG can also effectively improve the sample utilization rate, thereby improving the rationality of the task offloading strategy output by the task offloading processing model obtained through training.

[0172] Among them, the above-mentioned modules in the device provided by the embodiment of the present application can also implement the method steps provided by the embodiment of the training method of the above-mentioned task offloading processing model. Alternatively, the device provided by the embodiment of the present application can also include other modules in addition to the above-mentioned modules to implement the method steps provided by the embodiment of the training method of the above-mentioned task offloading processing model. And the device provided by the embodiment of the present application can achieve the technical effects that can be achieved by the embodiment of the training method of the above-mentioned task offloading processing model.

[0173] Among them, the above-mentioned modules in the device provided by the embodiment of the present application can also implement the method steps provided by the embodiment of the task offloading processing method for passive devices in the edge network. Alternatively, the device provided by the embodiment of the present application can also include other modules in addition to the above-mentioned modules to implement the method steps provided by the embodiment of the task offloading processing method for passive devices in the edge network. And the device provided by the embodiment of the present application can achieve the technical effects that can be achieved by the embodiment of the task offloading processing method for passive devices in the edge network.

[0174] Preferably, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and runnable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned task offloading processing method and model training method embodiments are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0175] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, each process of the above-mentioned task offloading processing method and model training method embodiment is implemented, and the same technical effect is achieved. To avoid repetition, it is not repeated here. Wherein, the computer-readable storage medium is such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0176] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program can be operated to enable a computer to execute part or all of the steps of the above-mentioned task offloading processing method and model training method embodiments, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0177] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0178] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0179] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0180] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0181] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0182] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0183] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media 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 memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0184] It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0185] Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer readable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon.

[0186] The embodiments of the present application described above are intended to be merely exemplary and those skilled in the art shall understand that modifications and variations can be made to the embodiments without departing from the spirit and scope of the application. Accordingly, the application is not limited to the embodiments described above.

Claims

1. A task offloading processing method, characterized in that: Host devices used in edge networks include: Obtain the processing demand information corresponding to multiple task nodes in the task flow to be processed and the available resource information of each device in the edge network; Perform feature coding on the processing requirement information and the available resource information based on feature dimensions of a preset state space to obtain task state features corresponding to multiple task nodes and environment features corresponding to each device; The task state characteristics corresponding to the plurality of task nodes and the environmental characteristics corresponding to the various devices are input into a pre-trained task offloading processing model to obtain a task offloading strategy corresponding to the task process to be processed, wherein the task offloading strategy includes a correspondence between the task nodes to be offloaded and the target offloading devices, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with the training goal of shortening task processing time, and the training data includes sample task processing time and sample state characteristics and sample environmental characteristics corresponding to the sample task in the preset state space; Unloading the task node to be unloaded to the corresponding target unloading device for execution based on the task unloading strategy; A task processing result corresponding to the task flow to be processed is generated based on the task node processing result fed back by the target offloading device.

2. The method according to claim 1, wherein Obtain the processing requirement information corresponding to multiple task nodes in the task flow to be processed and the available resource information of each device in the edge network, including: Compile the task flow to be processed into a directed acyclic graph through a task compiler, wherein the directed acyclic graph includes multiple task nodes connected by directed lines, and each task node carries corresponding processing requirement information; The available resource information of each device in the edge network is acquired through the global resource broadcast of the edge network.

3. A task offloading processing method, characterized in that: Passive devices used in edge networks include: receiving a task node offloaded by a host device in an edge network based on a task offloading strategy, wherein the task offloading strategy is obtained by the host device inputting task state characteristics corresponding to a plurality of task nodes in a task flow to be processed and environmental characteristics corresponding to each device in the edge network into a pre-trained task offloading processing model, wherein the task offloading processing model is obtained by training a value memory graph model based on training data with a training goal of shortening task processing time, and the training data includes sample task processing time and sample state characteristics and sample environmental characteristics corresponding to the sample task in a preset state space; Processing the task node based on locally available resources to obtain a task node processing result; Feedback the task node processing result to the host device.

4. The method according to claim 3, wherein Before processing the task node based on locally available resources, the method further includes: When there are multiple pending task nodes received, the congestion manager sequentially allocates the pending task nodes to locally available virtual machines; Feedback of the task node processing result to the host device includes: The congestion manager feeds back the processing result of the local available virtual machine on the task node to be processed to the host device.

5. A training method for a task offloading processing model, characterized in that: include: Obtain a sample of a task process to be processed and an edge network environment sample, wherein the sample of the task process to be processed includes multiple sample task nodes, each of which corresponds to processing requirement information of a sample task, and the edge network environment sample includes available resource information of multiple sample devices in the edge network; Perform feature coding on the processing requirement information and the available resource information based on the feature dimension of the preset state space to obtain sample state features corresponding to multiple sample task nodes and sample environment features corresponding to each sample device; Constructing a Markov decision process model with a graph structure corresponding to the sample environment characteristics through a value memory graph algorithm; Based on the sample state characteristics and sample task processing time corresponding to a plurality of sample task nodes, the Markov decision process model is trained with the goal of shortening the task processing time to obtain the task offloading processing model.

6. The method according to claim 5, wherein Based on the sample state characteristics and sample task processing time corresponding to a plurality of sample task nodes, the Markov decision process model is trained with shortening the task processing time as a training goal to obtain the task offloading processing model, including: Based on the sample state characteristics and sample task processing time corresponding to multiple sample task nodes, the training target is adjusted through a proximal strategy optimization algorithm, and the Markov decision process model is iteratively trained to obtain the task offloading processing model.

7. A task offloading processing system, characterized in that: comprising a plurality of devices connected in communication based on an edge network, wherein the plurality of devices include at least one host device and at least one passive device; The host device is used to implement the steps of the method according to claim 1 or 2; The passive device is used to implement the steps of the method according to claim 3 or 4.

8. The system according to claim 7, wherein: The host device includes a vehicle-mounted terminal, and the passive device includes a mobile terminal directly connected to the vehicle-mounted terminal for communication.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method according to any one of claims 1 to 6 when executed by the processor.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images