A method for predicting and offloading dynamically divisible tasks in the Internet of Vehicles scenario
By adopting the TPITOA algorithm and LSTM/DQN algorithm in the Internet of Vehicles to optimize task offloading decisions, the task offloading problem in the dynamic traffic environment of the Internet of Vehicles is solved, and resource utilization and processing efficiency are improved.
Patent Information
- Application Number
- CN202410891035.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2044-07-04
AI Technical Summary
In the Internet of Vehicles, traditional edge computing architecture cannot effectively cope with dynamically changing traffic flows and complex traffic environments, resulting in irrational task offloading decisions, low resource utilization, and extended processing time.
The TPITOA algorithm is adopted, combined with LSTM and deep reinforcement learning (DQN) algorithms. By predicting future traffic flow, the task offloading sequence is optimized using the methods of giving priority to the most intense ones, giving priority to the minimum number of alternative nodes, and giving priority to the minimum data volume and computing ratio, and the appropriate edge computing nodes are selected for task offloading.
It reduces the average processing time of tasks, improves resource utilization and processing efficiency, optimizes service quality, and reduces system resource consumption.
Smart Images
Figure CN118860646B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of optimized scheduling, and in particular to a method for predicting and unloading dynamically divisible tasks in a vehicle networking scenario. Background Art
[0002] The Internet of Vehicles (IoV) is a high-tech, modern transportation system designed to improve the safety, efficiency, and environmental friendliness of road transport. Using wireless communications, sensor technology, and big data analytics, IoV enables real-time monitoring and management of vehicle operating status and road conditions. These technologies enable vehicles to exchange information in real time, such as traffic flow, accidents, or road construction, enabling drivers to make faster and more accurate decisions. IoV can also help optimize vehicle routes and energy use, improving fuel efficiency and reducing emissions. Furthermore, IoV supports the development of autonomous driving technology, improving road efficiency and reducing traffic accidents through precise vehicle control. Overall, IoV reflects the modern trend in the transportation industry, promoting a more efficient, safer, and environmentally friendly transportation system.
[0003] Edge computing is an advanced computing architecture designed to optimize information processing and response times by geographically distributing data. It offloads data processing tasks from centralized data centers to edge devices near the data source, enabling real-time data processing and analysis. This computing approach can significantly reduce data transmission latency, improving system response speed and operational efficiency. Edge computing has a wide range of applications, including smart cities, the Industrial Internet, the Internet of Vehicles, and the Internet of Things. In the Internet of Vehicles, edge computing can process the massive amounts of data generated by onboard sensors, supporting real-time driving decisions and road condition analysis. Furthermore, edge computing can reduce data processing costs, reduce reliance on central data centers, and enhance data security and privacy. By performing preliminary processing at the data generation site and sending only necessary data to the cloud, it effectively reduces the amount of data transmitted and alleviates network burden. Overall, edge computing represents a new trend in computing technology, providing faster, more efficient, and more secure services by processing data closer to the user.
[0004] Mobile Edge Computing (MEC) is a key component of the edge computing architecture of the Internet of Vehicles (IoV) and the 5G architecture. MEC moves computing, caching, and networking functions to the edge of the network, providing data, storage, and computing services to user vehicles. Compared to traditional cloud computing solutions, edge-based MEC solutions can both provide the computing power required for compute-intensive tasks and reduce the latency associated with transmission tasks, achieving low latency and low bandwidth consumption, thereby improving the QoS of mobile applications. Task offloading is a key feature of MEC technology, which refers to the offloading of computing tasks from terminal devices to nearby edge devices or to collaborative processing between edge and cloud devices. Due to the additional energy consumption and time loss associated with data transmission and remote task execution, task offloading and resource caching are research issues in edge computing. Considering the nodes for task offloading, task offloading decisions, and task resource allocation, it is necessary to consider the resource availability of edge nodes, task resource consumption, and transmission channel conditions to arrive at a reasonable offloading decision.
[0005] In the context of edge computing in the connected vehicle (IoV), the dynamic flow of traffic, high-speed vehicles, and complex road conditions make traditional edge computing architectures impractical for direct application. Because traffic volume within a region reflects the number of requests with varying degrees of severity, the number of service requests varies with traffic volume. Therefore, in the complex traffic environment unique to the IoV, dynamic traffic flow has a direct impact on task offloading and resource allocation within the IoV. Task offloading decisions are adjusted based on real-time traffic conditions. In an in-vehicle edge computing environment, the task processing of intelligent vehicle applications is complex and diverse, encompassing not only simple independent 0-1 tasks but also tasks such as sensor data processing and vehicle status monitoring, which can be processed in parallel. For example, autonomous driving systems involve real-time collection and analysis of vehicle sensor data, continuous vehicle status monitoring, dynamic path planning and navigation, and real-time interaction with the external environment. For environmental monitoring applications, these analytical tasks are independent of each other. Although the data processed by each task comes from the same data source (vehicle sensor data), each task can be independently processed locally or offloaded to a remote server, and the processing results are not dependent on each other.
[0006] Nam et al. [1] proposed a new edge computing architecture and task offloading scheme, using RSUs as edge computing nodes to support automated vehicle platoon tasks. RSUs provide low-latency computing services for platoon tasks at the location closest to the vehicle. Li et al. [2] considered the problem of sequentially distributing splittable applications across multiple processor nodes with limited buffer sizes and proposed an incremental balancing strategy for achieving optimal load distribution. However, the above studies did not analyze the dynamic and splittable task aspects of the Internet of Vehicles.
[0007] [1] S.Nam, S.Kwak, J.Lee, and S.Park, "Task offloading based on vehicular edge computing for autonomous platooning," Comput.Syst.Sci.Eng., vol.46, no.1, pp.659–670, 2023.
[0008] [2] Summary of the Invention
[0009] To solve the above problems, the present invention discloses a method for dynamically divisible task prediction and offloading in the Internet of Vehicles scenario. The present invention mainly considers the problem of dynamic task prediction and offloading in the edge computing of the Internet of Vehicles from the perspective of minimizing task processing time, fully considering the characteristics of traffic flow, vehicle mobility, simple independent tasks, and divisible tasks in the Internet of Vehicles scenario and their respective impacts, and uses the TPITOA algorithm to optimize the target problem, thereby minimizing the average task processing time.
[0010] The technical solution adopted in the present invention is:
[0011] A method for predicting and offloading dynamically divisible tasks in an Internet of Vehicles scenario includes the following stages:
[0012] A preprocessing module: Builds a divisible task offloading scheduling system in an IoV with multiple edge computing nodes. Selects an edge computing node as a centralized scheduler to make decisions on offloading tasks carried by IoV vehicle users. It also builds resource, task, computing, transmission, and prediction models to monitor the overall system status. The remaining computing nodes are used for computing and offloading functions.
[0013] B. Offloading sequence sorting module: When a computing node receives a task computing request, it sends the task attribute information to the scheduler. The centralized scheduler then determines the task offloading sequence by prioritizing the most intense competitor, the least number of candidate nodes, and the smallest computing amount and data volume ratio, optimizing the sequence sorting and thus optimizing the system's processing efficiency and resource utilization.
[0014] C offloading decision module: The centralized scheduler uses an independent and divisible task offloading scheduling method based on traffic flow prediction. The scheduler updates the state information of the environment, makes action decisions in the action space, offloads tasks to the most appropriate nodes, and optimizes decision efficiency by setting the reward function.
[0015] D Decision-making module: The centralized scheduler sends the decision to the computing node, and the edge computing node performs local calculation or forwarding on the task request according to the scheduling result.
[0016] The specific steps of the pre-processing module are as follows:
[0017] A1. Build a task offloading scheduling control system in edge nodes. Select an edge computing node with the most remaining computing resources as the centralized scheduler to make decisions on task offloading scheduling.
[0018] A2. The centralized scheduler builds resource, task, computation, transmission, and prediction models, and monitors the computing resource utilization and memory resource utilization of all computing nodes, the status of all edge servers, and the execution status of all tasks.
[0019] A3. The remaining computing nodes can accept task requests from vehicle users, execute task requests, or offload vehicle user computing tasks to other computing nodes. Tasks can also be processed locally.
[0020] The uninstall sequence sorting module includes the following steps:
[0021] B1. When a computing node receives a task execution request, it sends the data size and task computational load information to the scheduler.
[0022] B2. The centralized scheduler sorts the tasks generated by vehicle users by giving priority to the most intense ones, the smallest number of alternative nodes, and the smallest data volume and computing volume ratio.
[0023] The uninstall decision module includes the following steps:
[0024] C1. The centralized scheduler sends the decision to the compute nodes.
[0025] C2. The centralized scheduler offloads tasks to the most appropriate node using an independent and divisible task offloading scheduling method based on traffic flow prediction. Regarding the two basic methods of independent and divisible task offloading scheduling based on traffic flow prediction, the LSTM prediction method processes the historical traffic flow data of each edge computing node to predict the traffic flow conditions of each edge computing node within a certain period of time in the future. This data serves as the environmental input for deep reinforcement learning (DQN). Through iterative updates of the state space, action space, and reward function, the advantages and disadvantages of each possible offloading method are determined.
[0026] C3. The edge computing node offloads the task to the designated node based on the decision, and the computing resources of the designated node are used for calculation. If the designated decision target is the vehicle user, the calculation is performed locally. If the designated node has insufficient remaining resources, the task sequence is unloaded and waits.
[0027] The B2 workflow includes the following steps:
[0028] S1. The centralized scheduler receives all vehicle-carried tasks in the current phase, first classifies and sorts them by vehicle, and then sorts the tasks generated by vehicle users according to the method of prioritizing the most intense tasks, the smallest number of candidate nodes, and the smallest data volume and computational load ratio. After selecting a specific method, the process proceeds to S2.
[0029] S2. Prioritize the most intense candidate and go to S3; prioritize the candidate with the smallest number of candidate nodes and go to S4; prioritize the candidate with the smallest data volume and computing volume ratio and go to S5;
[0030] S3. The most competitive priority rule considers all competitors for each task and the available resources to determine the intensity of competition. Taking into account the number of requested tasks and available resources on the optional service nodes for each task, the task with the most intense competition has a higher priority;
[0031] S4. The priority rule for the minimum number of candidate nodes is that tasks with fewer candidate RSUs are given priority;
[0032] S5. Select the minimum data volume and computing volume ratio priority rule, calculate the ratio of the data volume carried by the task and the computing volume required, and give priority to executing the task with the smaller ratio.
[0033] S6. Return the sorted tasks to the centralized scheduler to prepare for the next stage of work.
[0034] The C2 workflow consists of the following steps:
[0035] S1. Traverse and check whether each computing node can accept offloading, thereby obtaining a list of nodes that can accept task offloading;
[0036] S2. Determine whether the list of nodes that can accept task offloading is empty. If so, go to S4; otherwise, go to S3.
[0037] S3 determines whether the vehicle user has local execution capabilities. If so, go to S4; otherwise, go to S5;
[0038] S4. Select the vehicle local node calculation function;
[0039] S5. Select a suitable node for task offloading;
[0040] S6. After selecting a suitable node, determine whether the node is a local node. If so, go to S4; otherwise, go to S7;
[0041] S7. Unload the task data to the designated computing node;
[0042] S8. Determine whether the node that needs to calculate the vehicle splittable task has idle computing resources. If so, go to S9; otherwise, go to S10;
[0043] S9. Select the amount of computing resources required for the task corresponding to the node, calculate the task, and transmit the result back to the vehicle user after the task is completed;
[0044] S10. Arrange the offload tasks on the node, wait for resources to be released, and execute the computing tasks after sufficient resources are released.
[0045] The step S1 comprises the following steps:
[0046] S11. Use the historical traffic flow data of each computing node to predict the traffic flow data at future times, go to S12, combine the current traffic flow data, determine the number of tasks in the scenario, obtain the task offloading decision, and go to S13.
[0047] S12. Use the LSTM method to select a certain number of historical traffic flow values and predict traffic flow data at future moments;
[0048] S13. Based on the predicted traffic value, a deep reinforcement learning (DQN) algorithm is used to iteratively update the environment state space, action space, and reward function. Based on the reward value derived from the reward function, reasonable task offloading decisions and the split ratio of divisible tasks are derived.
[0049] S14. Based on the obtained action decision, simple independent tasks are offloaded to the corresponding computing nodes. For divisible tasks, the corresponding task parts are offloaded to the corresponding computing nodes according to the split ratio, and finally an offloading decision list is obtained.
[0050] A device for implementing vehicle-mounted edge computing task offloading, comprising:
[0051] processor;
[0052] a memory storing executable instructions of the processor and storing historical traffic flow data;
[0053] Wherein, the processor is configured to execute the steps of the implementation method of offloading the on-board edge computing task.
[0054] A storage medium that can be recognized by an edge server is used to store historical traffic flow data and a program. When the program is executed, the steps of the method for implementing the vehicle-at-edge computing task offloading are implemented.
[0055] Beneficial effects of the present invention: Compared with the prior art, the method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario provided by the present invention has the following beneficial effects:
[0056] 1. This invention is the first method to consider the characteristics of vehicle dynamics and divisible tasks in the edge computing scenario of the Internet of Vehicles. Taking into account the dynamic changes in the traffic environment of the Internet of Vehicles,
[0057] 2. The present invention estimates the road traffic information within a certain period of time in the future, so that it can select more efficient offloading decisions, thereby reducing the average processing time of tasks.
[0058] 3. When making reasonable task offloading decisions, this invention considers offloading divisible tasks to more appropriate computing nodes based on the split ratio determined by the decision, thereby avoiding resource exhaustion and increased processing time. Furthermore, for the predicted traffic data used in the algorithm, we utilize an LSTM algorithm to obtain future traffic data, ensuring better offloading decisions.
[0059] 4. For the problem of cold divisibility tasks, due to its particularity to task characteristics, it fully utilizes the ability of multiple edge nodes to process tasks simultaneously, reducing the average processing time of tasks and thus increasing resource utilization.
[0060] 5. Therefore, the present invention comprehensively considers the characteristics of the vehicle network scene status and the vehicle-carrying tasks, optimizes service quality, reduces system resource consumption, and improves task processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 This is a schematic diagram of task offloading in the Internet of Vehicles edge computing scenario of the present invention.
[0062] Figure 2 This is a schematic diagram of the application of edge computing to the Internet of Vehicles in the present invention.
[0063] Figure 3This is a detailed flow chart of the dynamic divisible task prediction and offloading method based on the Internet of Vehicles scenario in the present invention.
[0064] Figure 4 This is a detailed flow chart of obtaining a list of nodes that can accept task offloading in the present invention. DETAILED DESCRIPTION
[0065] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, and the terms "inward" and "outward" refer to directions toward or away from the geometric center of a particular component, respectively.
[0066] Example 1
[0067] A method for predicting and offloading dynamically divisible tasks in an Internet of Vehicles scenario includes the following stages:
[0068] A. Preprocessing Module: This module builds a divisible task offloading scheduling system in an IoV with multiple edge computing nodes. A specific edge computing node is selected as the centralized scheduler to make decisions on offloading tasks carried by IoV vehicle users. It also builds resource, task, computation, transmission, and prediction models to monitor the overall system status. The remaining computing nodes are used for computation and offloading functions.
[0069] B. Offloading sequence sorting module: When a computing node receives a task computing request, it sends the task attribute information to the scheduler. The centralized scheduler then determines the task offloading sequence by prioritizing the most intense competitor, the least number of candidate nodes, and the smallest computational workload and data volume ratio, optimizing the sequence and thus optimizing the system's processing efficiency and resource utilization.
[0070] C. Offloading decision module: The centralized scheduler uses an independent and divisible task offloading scheduling method based on traffic flow prediction. The scheduler updates the state information of the environment, derives the action decision of the action space, offloads the task to the most appropriate node, and optimizes the decision efficiency by setting the reward function.
[0071] D. Decision-making module: The centralized scheduler sends the decision to the computing node.
[0072] The edge computing node performs local calculations or forwards the task request according to the scheduling results.
[0073] The specific steps of the pre-processing module are as follows:
[0074] A1. Build a task offloading scheduling control system in edge nodes. Select an edge computing node with the most remaining computing resources as the centralized scheduler to make decisions on task offloading scheduling.
[0075] A2. The centralized scheduler builds resource, task, computation, transmission, and prediction models, and monitors the computing resource utilization and memory resource utilization of all computing nodes, the status of all edge servers, and the execution status of all tasks.
[0076] A3. The remaining compute nodes can accept task requests from vehicle users, execute them, or offload the user's computing tasks to other compute nodes. Tasks can also be processed locally. These nodes include: ES1, ES2, ES3, ES4, ES5, and ES6. Node attributes include computing power and coverage. Nodes are geographically fixed, with varying but fixed coverage areas. They communicate with vehicles via V2I.
[0077] The uninstall sequence sorting module includes the following steps:
[0078] B1. When a computing node receives a task execution request, it sends the data size and task computational load information to the scheduler.
[0079] B2.ES i When a function computing request is received, the data size of the task and the amount of computing required for the task are sent to the scheduler; the centralized scheduler sorts the tasks generated by vehicle users by giving priority to the most intense ones, the smallest number of alternative nodes, and the smallest data volume and computing amount ratio.
[0080] B3. As shown in the unloading sequence arrangement process below, the algorithm obtains the final sorting result.
[0081] The uninstall decision module includes the following steps:
[0082] C1. The centralized scheduler sends the decision to the compute nodes.
[0083] C2. The centralized scheduler offloads tasks to the most appropriate node using an independent and divisible task offloading scheduling method based on traffic flow prediction. Regarding the two basic methods of independent and divisible task offloading scheduling based on traffic flow prediction, the LSTM prediction method processes the historical traffic flow data of each edge computing node to predict the traffic flow conditions of each edge computing node within a certain period of time in the future. This data serves as the environmental input for deep reinforcement learning (DQN). Through iterative updates of the state space, action space, and reward function, the advantages and disadvantages of each possible offloading method are determined.
[0084] C3. The edge computing node offloads the task to the designated node based on the decision, and the computing resources of the designated node are used for calculation. If the designated decision target is the vehicle user, the calculation is performed locally. If the designated node has insufficient remaining resources, the task sequence is unloaded and waits.
[0085] The B2 workflow includes the following steps:
[0086] S1. The centralized scheduler receives all vehicle-carried tasks in the current phase, first classifies and sorts them by vehicle, and then sorts the tasks generated by vehicle users according to the method of prioritizing the most intense tasks, the smallest number of candidate nodes, and the smallest data volume and computational load ratio. After selecting a specific method, the process proceeds to S2.
[0087] S2. Prioritize the most intense candidate and go to S3; prioritize the candidate with the smallest number of candidate nodes and go to S4; prioritize the candidate with the smallest data volume and computing volume ratio and go to S5;
[0088] S3. The most competitive priority rule considers all competitors for each task and the available resources to determine the intensity of competition. Taking into account the number of requested tasks and available resources on each task's optional service nodes, the most competitive task is given priority. The specific variable parameters involved in this rule are as follows:
[0089]
[0090] According to the rules, the tasks with the most intense competitors are uninstalled first.
[0091] S4. The rule with the minimum number of candidate nodes is to prioritize tasks with fewer candidate RSUs. An example of this rule is as follows:
[0092] Task Optional nodes Priority selection <![CDATA[T1]]> <![CDATA[ES1,ES3,ES4,ES5]]> no <![CDATA[T2]]> <![CDATA[ES1,ES2]]> yes
[0093] S5. Select the minimum data volume and computational load ratio priority rule. Calculate the ratio between the amount of data carried by the task and the computational load required, and prioritize the task with the smaller ratio. An example of this rule is as follows:
[0094] Task Task data volume Calculate demand RDC value Priority selection <![CDATA[T1]]> 160MB 100MB 0.1875 yes <![CDATA[T2]]> 220MB 180MB 0.2455 no
[0095] S6. Return the sorted tasks to the centralized scheduler to prepare for the next stage of work.
[0096] The C2 workflow consists of the following steps:
[0097] S1. Traverse and check whether each computing node can accept offloading, thereby obtaining a list of nodes that can accept task offloading;
[0098] S2. Determine whether the list of nodes that can accept task offloading is empty. If so, go to S3; otherwise, go to S5.
[0099] S3. Determine whether the vehicle user has local execution capability. If so, go to S4;
[0100] S4. Select the vehicle local node computing task;
[0101] S5. After selecting the appropriate node, determine whether the node is a local user. If so, go to S4; otherwise, go to S6;
[0102] S6. Unload the task data to the designated computing node;
[0103] S7. Determine whether the node that needs to calculate the vehicle splittable task has idle computing resources. If so, go to S8; otherwise, go to S9;
[0104] S8. Select the amount of computing resources required for the task corresponding to the node, calculate the task, and transmit the result back to the vehicle user after the task is completed;
[0105] S9. Arrange the offload tasks on the node, wait for resources to be released, and execute the computing tasks after sufficient resources are released.
[0106] The step S1 comprises the following steps:
[0107] S11. Use the historical traffic flow data of each computing node to predict the traffic flow data at future times, go to S12, combine the current traffic flow data, determine the number of tasks in the scenario, obtain the task offloading decision, and go to S13.
[0108] S12. Use the LSTM method to select a certain number of historical traffic flow values and predict traffic flow data at future moments;
[0109] S13. Based on the predicted traffic value, a deep reinforcement learning (DQN) algorithm is used to iteratively update the environment state space, action space, and reward function. Based on the reward value derived from the reward function, reasonable task offloading decisions and the split ratio of divisible tasks are derived.
[0110] S14. Based on the obtained action decision, simple independent tasks are offloaded to the corresponding computing nodes. For divisible tasks, the corresponding task parts are offloaded to the corresponding computing nodes according to the split ratio, and finally an offloading decision list is obtained.
[0111] Through the above process, the present invention implements a method for predicting and offloading dynamically divisible tasks in the Internet of Vehicles scenario. By using the LSTM traffic prediction method in conjunction with the deep reinforcement learning algorithm (DQN), the dynamic nature of the Internet of Vehicles scenario is grasped, reasonable offloading decisions are made, and the offloading ratio and offloading target of the task are obtained. Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and purpose of the present invention. The scope of the present invention is defined by the claims and their equivalents.
[0112] The technical means disclosed in the solution of the present invention are not limited to the technical means disclosed in the above-mentioned embodiment, but also include technical solutions composed of any combination of the above technical features.
Claims
1. A method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario, characterized by: The following stages are included: A. Preprocessing stage: A divisible task offloading scheduling system is built in the IoV with multiple edge computing nodes. An edge computing node is selected as the centralized scheduler to make decisions on the offloading scheduling of tasks carried by vehicle users in the IoV. Resource, task, computing, transmission, and prediction models are built to monitor the overall status of the system. The remaining computing nodes are used for computing and offloading functions. B. Offloading sequence sorting stage: When a computing node receives a task computing request, it sends the task attribute information to the scheduler. The centralized scheduler then determines the task offloading sequence by prioritizing the most intense competitor, the least number of candidate nodes, and the smallest computing amount and data volume ratio, optimizing the sequence sorting and thus optimizing the system's processing efficiency and resource utilization. C. Offloading decision stage: The centralized scheduler uses an independent and divisible task offloading scheduling method based on traffic flow prediction. The scheduler updates the state information of the environment, makes action decisions in the action space, offloads tasks to the most appropriate nodes, and optimizes decision efficiency by setting the reward function. The centralized scheduler offloads tasks to the most appropriate nodes using an independent and divisible task offloading scheduling method based on traffic flow prediction. Regarding the two basic methods for independent and divisible task offloading scheduling based on traffic flow prediction, the LSTM prediction method processes historical traffic flow data from each edge computing node to predict traffic flow conditions for each edge computing node within a certain period of time in the future. This data serves as the input to the deep reinforcement learning environment. Through iterative updates of the state space, action space, and reward function, the advantages and disadvantages of each possible offloading method are determined. D Decision-making stage: The centralized scheduler sends the decision to the computing node, and the edge computing node performs local calculation or forwarding on the task request according to the scheduling result.
2. The method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario according to claim 1, characterized in that: A preprocessing stage specifically includes the following steps: A1. Build a task offloading scheduling control system in edge nodes. Select an edge computing node with the most remaining computing resources as the centralized scheduler to make decisions about task offloading scheduling. A2. The centralized scheduler builds resource, task, computation, transmission, and prediction models, and monitors the computing resource utilization and memory resource utilization of all compute nodes, the status of all edge servers, and the execution status of all tasks. A3. The remaining computing nodes can accept task requests from vehicle users, execute task requests, or offload vehicle user computing tasks to other computing nodes. Tasks can also be processed locally.
3. The method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario according to claim 1, characterized in that: B Uninstall sequence sorting phase specifically includes the following steps: B1. When a compute node receives a task execution request, it sends the data size and task computational complexity to the scheduler. B2. The centralized scheduler prioritizes the most intense tasks, the smallest number of candidate nodes, and the smallest data volume and computational load ratio to prioritize the tasks generated by vehicle users.
4. The method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario according to claim 1, characterized in that: The C uninstall decision phase specifically includes the following steps: C1. The centralized scheduler sends the decision to the compute nodes. C2. The edge computing node offloads the task to the designated node based on the decision, and the computing resources of the designated node are used for calculation. If the designated decision target is the vehicle user, the calculation is performed locally. If the designated node has insufficient remaining resources, the offload task sequence is waited.
5. The method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario according to claim 4, characterized in that: Step C2 specifically includes the following steps: S1. Traverse and check whether each compute node can accept offloading, thereby obtaining a list of nodes that can accept task offloading. S2. Determine whether the list of nodes that can accept task offloading is empty. If so, proceed to S4. Otherwise, go to S3; S3. Determine whether the vehicle user has local execution capabilities. If so, proceed to S4. Otherwise, go to S5; S4. Select the vehicle local node calculation function; S5. Select a node with sufficient resources, meeting the task deadline constraints, and ensuring good load balancing over time for task offloading. S6. After selecting a suitable node, determine whether the node is a local node. If so, go to S4; otherwise, go to S7. S7. Offload the task data to the specified computing node. S8 determines whether the node that needs to calculate the vehicle splittable task has idle computing resources. If so, go to S9; otherwise, go to S10; S9. Select the amount of computing resources required for the task corresponding to the node, calculate the task, and transmit the results back to the vehicle user after the task is completed; S10. Arrange the offload tasks on the node and wait for resources to be released. After sufficient resources are released, execute the computing tasks.
6. The method for predicting and offloading dynamically divisible tasks in a connected vehicle scenario according to claim 5, characterized in that: Step S5 specifically includes the following steps: S11. Use historical traffic flow data from each computing node to predict future traffic flow data. Then, proceed to S12. Combined with current traffic flow data, determine the number of tasks in the scenario, make a task offloading decision, and proceed to S13. S12. Use the LSTM method to select a certain number of historical traffic flow values and predict traffic flow data at future times. S13. Based on the predicted traffic values, a deep reinforcement learning algorithm is used to iteratively update the environment state space, action space, and reward function. Based on the reward value derived from the reward function, reasonable task offloading decisions and the split ratio of divisible tasks are determined. S14. Based on the obtained action decision, simple independent tasks are offloaded to the corresponding computing nodes. For divisible tasks, the corresponding task parts are offloaded to the corresponding computing nodes according to the split ratio, and finally an offloading decision list is obtained.
7. A device for implementing vehicle-mounted edge computing task offloading, characterized in that: include: processor; a memory storing executable instructions of the processor and storing historical traffic flow data; The processor is configured to execute the steps of a method for dynamically divisible task prediction and offloading in a vehicle networking scenario according to any one of claims 1 to 6 by executing the executable instructions.
8. A storage medium identifiable by an edge server for storing historical traffic flow data and programs, characterized in that: When the program is executed, the steps of a method for predicting and offloading dynamically divisible tasks in a vehicle networking scenario according to any one of claims 1 to 6 are implemented.