Edge task unloading strategy of vehicle infrastructure cooperative system fused with feedback link perception
By constructing a refined task latency model and the DGA algorithm, which combines Dijkstra's multi-strategy generation algorithm with a genetic algorithm, the task offloading path and equipment allocation of the vehicle-road cooperative system are optimized, solving the latency and resource shortage problems caused by high-complexity tasks and achieving low-latency and high-success-rate task offloading.
Patent Information
- Application Number
- CN202510941763.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-10-28
AI Technical Summary
In existing vehicle-road cooperative systems, as the number of highly complex tasks increases, the large amount of computational data leads to non-negligible transmission latency, and the system suffers from insufficient resources during peak hours. Existing research has failed to effectively solve the problems of task unloading delay and success rate.
This paper proposes an edge task offloading strategy that integrates feedback link perception in a vehicle-road cooperative system. By constructing a refined task delay model and using the DGA algorithm that combines Dijkstra's multi-policy generation algorithm with a genetic algorithm, the task offloading path and device allocation are optimized, thereby reducing latency and improving success rate.
It effectively reduces task unloading delay, improves system efficiency, enhances task success rate, optimizes resource allocation, and improves system stability and iteration speed.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention belongs to the field of edge computing and wireless communication technology, and specifically relates to an edge task offloading strategy for a vehicle-road cooperative system that integrates feedback link perception. Background Technology
[0002] In recent years, the Internet of Things (IoT) has become an important research field. In IoT systems, people, data, and objects can connect to each other and access the internet. Vehicles, as a key component of the IoT, have become the third most frequented place for people outside of their homes and offices. Most people tend to use smartphones and tablets inside their vehicles. Meanwhile, with the continuous advancement of vehicle technology, more and more new application scenarios are emerging, requiring real-time computing and data processing capabilities to meet the operational needs of vehicles. Therefore, the application of Vehicular Edge Computing (VEC) technology is becoming increasingly widespread, becoming an important technology supporting vehicle-to-everything (V2X) and other IoT scenarios. VEC achieves distributed offloading of computing and storage resources by offloading computationally intensive tasks from terminal devices to edge nodes. This process effectively reduces task latency and communication burden; compared to traditional cloud computing, VEC has lower deployment costs and better real-time response capabilities.
[0003] In VEC, there are two main task offloading modes. The first is Vehicle-to-Infrastructure (V2I) communication, where vehicles obtain data access services through Roadside Units (RSUs). The second is Vehicle-to-Vehicle (V2V) communication, where vehicles can communicate directly with each other via IEEE 802.11p or C-V2V modes. Through these wireless transmission technologies, vehicles can pass tasks to mobile edge servers or other vehicles for processing.
[0004] Current research focuses on task offloading strategies in VEC (Vehicle-Road Cooperative) systems. C. Chen et al. (see C. Chen, Y. Zeng, H. Li, Y. Liu and S. Wan, "A Multihop Task Offloading Decision Model in MEC-Enabled Internet of Vehicles," IEEE Internet of Things Journal, vol. 10, no. 4, pp. 3215-3230, 2023.) developed a distributed multihop task offloading decision efficiency model. This model can offload customer vehicle tasks to suitable service vehicles outside the communication range by calculating the path connection time and path transmission time between vehicles, reducing task processing latency and thus improving task completion efficiency. L. Zhao et al. (“MESON: A Mobility-Aware Dependent Task Offloading Scheme for Urban Vehicular Edge Computing,” IEEE Transactions on Mobile Computing, vol. 23, no. 5, pp. 4259–4272, 2024.) established a computational offloading model by considering task data dependencies, trained the offloading strategy using the DRL algorithm, and designed a vehicle mobility detection algorithm and a task priority determination algorithm. This effectively reduced the average response time and average energy consumption, and improved system stability and offloading success rate.
[0005] In VEC networks, with the rapid increase in the number of highly complex tasks, the amount of computational results data is often large, making the return latency non-negligible. Simultaneously, the large amount of data stacking leads to insufficient computational resources for RSUs during peak system periods, and offloading tasks to resource-constrained RSUs may result in decreased system efficiency. However, the solutions proposed in previous studies have failed to effectively address these issues. Therefore, in a vehicle-road cooperative system with limited RSU resources, we fully leverage the advantages of multi-hop communication to significantly expand the target vehicle selection range. Our research fully considers the key impacts of vehicle mobility and task data volume on system performance and integrates these key factors with resource constraints into a unified optimization framework. Summary of the Invention
[0006] The purpose of this application is to overcome the shortcomings of existing technologies and propose an edge task offloading strategy that integrates feedback link perception in vehicle-road cooperative systems.
[0007] This application addresses a vehicle-to-infrastructure (V2I) cooperative system comprising N roadside units and M intelligent IoT vehicles, denoted as device S = U∪V. Within the system, devices issue tasks, which can be transmitted or computed by any device in the system via V2I and V2V communication. The device S is now defined as follows. i The r-th task issued is represented as l i,r The relevant attributes of its task are represented as follows These tasks will be offloaded to the appropriate devices via multi-hop transfer, within the limits of the computing capabilities and storage resources supported by the devices, thereby reducing the load on the original devices. The goal is to ensure the lowest possible average system latency while maintaining a high task success rate, while meeting the following conditions:
[0008] min αT average +(1-α)Q succ (1)
[0009] ∑ s∈S M ns =1, (2)
[0010] ∑ s∈S M ms =1, (3)
[0011]
[0012] Among these constraints, constraint (2) ensures that when a roadside unit publishes a task, it can only offload it to one service device. Constraint (3) ensures that when a vehicle publishes a task, it can only offload it to one service device, thus preventing the task from being published repeatedly. Constraint (4) ensures that the device transmitting the task must have sufficient storage space to receive the corresponding task, preventing the receiving device from exceeding its capacity. Constraint (6) ensures that the task type corresponds to the task processing capacity of the device, and the publisher device must offload the generated task to a service device with the capability to calculate the corresponding task type.
[0013] This invention proposes an edge task offloading strategy for a vehicle-road cooperative system that integrates feedback link perception, as detailed below:
[0014] 1. First, construct the network model of the vehicle-road cooperative system, and refine the specific time delays of the tasks, mainly dividing them into four parts: upload delay, waiting delay, computation delay, and feedback delay. In this process, unloading the task on the task publisher is called local unloading, and unloading on a non-task publisher is called non-local unloading. There is no upload delay or feedback delay in the local unloading process.
[0015] 1) For transmission latency, we divide it into upload latency and return latency. Upload latency represents the transmission time required for the task to be transmitted to the service device via the uplink, while return latency represents the time required for the task result to be transmitted back to the publishing device via the downlink. The transmission latency required for the same type of task will vary depending on the specific offloading link. Tasks that are locally offloaded are calculated on their own publisher device and do not require transmission or return, therefore they have no upload latency or return latency. Specifically, for tasks in non-local offloading processes, the upload latency is expressed as... The return latency is expressed as
[0016] 2) Computation latency is the time required for a task to be computed on the corresponding service device. This time is determined by the number of CPU cycles required for the task to be computed and the computing power of the service device itself. Specifically, it is expressed as:
[0017] 3) Latency is the time from when a task is transmitted to the service device to when it begins computation on that device. Tasks within the device are queued according to an M / M / 1 queue model. The number of tasks a device in the system can execute within a given time period is determined by its maximum computing resource limit D. s When the computing power of the service device is F s In this case, the waiting latency of a task can be quantified, specifically as follows:
[0018] 4) Based on the four types of delays obtained previously, we can determine the total delay of a single task's complete non-local unloading process, expressed as: Similarly, the total latency of a single task's complete non-local unloading process can be expressed as: Therefore, in a system with a total of H tasks, we can determine the maximum tolerable latency T within the system. max The overall task latency ratio of the system is obtained as follows:
[0019] 5) To further evaluate the efficiency of system task unloading, we set up a system with dimensions X... traffic ,Y traffic The study focuses on rectangular road networks, considering the maximum tolerable delay time of the task. The task is completed within the specified timeframe, and the uninstallation process is completed before the publisher leaves the road network, marked as "task successful". Therefore, we can calculate the task success rate model within a single road network of the system as follows: In vehicle-road cooperative systems, our goal is to minimize the average latency ratio while maximizing the task success rate.
[0020] 2. Based on the relative motion trajectories of vehicles and roadside units in the model, construct usable communication time links to determine stable communication conditions between vehicles and roadside units. This is done using any pair of intelligent connected vehicles (V-V) in the crossroads of the vehicle-road cooperative system. A With V B For example, the details are as follows:
[0021] 1) Based on the communication range and driving status of intelligent connected vehicles, define vehicle motion-related formulas, detailing them to angle steering and relative position changes, as follows:
[0022]
[0023] 2) Considering the intersection situation of vehicles, divided into straight-line travel and cross-traffic travel, the vehicle communication time is quantified, and the waiting delay caused by traffic lights is taken into account. Specifically, there are three categories: When car A and car B intersect, car A proceeds straight; when car A and car B intersect, car A merges into car B's lane traveling in the same direction; when car A and car B intersect, car A merges into car B's lane traveling in the opposite direction. Each category is then assigned a different... and The available communication time for the vehicle can be determined under all circumstances.
[0024] 3. The edge task offloading DGA strategy of the vehicle-road cooperative system fusion feedback link perception as described in requirement 1 has the same characteristics as the common NP-hard problem "flexible job shop scheduling problem under uncertain working hours". This is a more complex combinatorial optimization problem than traditional flexible job shop scheduling. The optimization problem can be solved by the DGA algorithm combining Dijkstra's multi-policy generation algorithm and genetic algorithm to determine the task offloading path and the overall task offloading strategy.
[0025] 1) Establish an intelligent connected vehicle communication network, generate the location of all devices in the road network at each time point, and store all information about the road network at each time point. Establish a task offloading strategy matrix M. i It is used to store all task unloading strategies and task unloading path plans in a single iteration simulation.
[0026] 2) Using Dijkstra's multi-strategy generation algorithm, the shortest current workshop distance is generated for all tasks. Based on the available communication time proposed in Requirement 2, the best unloading and transmission devices are allocated to the tasks to obtain the single-task unloading strategy matrix M in the current environment. x .
[0027] 3) Repeat the experiment multiple times to obtain the total task unloading strategy matrix N. j =[M1,…,M j ,…,M Y ], for each task, unload strategy matrix M iConstruct a chromosome, treating individual task unloading strategies as genes, and perform crossover and mutation operations, iterating repeatedly until the optimal task unloading strategy matrix M for the system is obtained. y The strategy represented by this matrix can meet the system requirements.
[0028] This invention demonstrates that by employing the above-mentioned strategy, the task offloading of the vehicle-road cooperative system can be effectively completed and the system utility can meet the required requirements.
[0029] Compared with the prior art, the beneficial effects of this invention are:
[0030] 1) The edge task offloading (DGA) strategy for vehicle-road cooperative systems proposed in this invention, which integrates feedback link perception, can rationally allocate task offloading devices and optimize task offloading paths based on vehicle movement trajectories and available communication time between vehicles, thereby reducing task offloading latency and task failure rate, and improving system efficiency (e.g., ...). Figure 3 (as shown);
[0031] 2) The edge task offloading DGA algorithm for vehicle-road cooperative systems with fusion feedback link perception proposed in this invention has better chain effect and impressive iteration speed compared to existing heuristic algorithms (e.g., Figure 4 (As shown). Attached Figure Description
[0032] Figure 1 This is a structural diagram of the intelligent connected vehicle-road cooperative system with edge server proposed in this invention.
[0033] Figure 2 This is a schematic diagram illustrating the iterative process using an elite strategy in a genetic algorithm.
[0034] Figure 3 This is a schematic diagram of the offloading path for the same task in a vehicle-road cooperative system under the DGA strategy and the existing strategy.
[0035] Figure 4 This is a schematic diagram comparing the fitness of the DGA algorithm and existing algorithms in a vehicle-road cooperative system with the number of iterations. Detailed Implementation
[0036] The edge task offloading strategy for vehicle-road cooperative systems that integrates feedback link awareness includes the following steps:
[0037] 1. First, construct the network model of the vehicle-road cooperative system, and refine the specific time delays of the tasks, mainly dividing them into four parts: upload delay, waiting delay, computation delay, and feedback delay. In this process, unloading the task on the task publisher is called local unloading, and unloading on a non-task publisher is called non-local unloading. There is no upload delay or feedback delay in the local unloading process.
[0038] 1) Upload latency represents the transmission time required for a task to be transmitted to the service device via the uplink, while return latency represents the time required for the task result to be transmitted back to the publishing device via the downlink. For non-local unloading processes, the upload latency is expressed as... The return latency is expressed as
[0039] 2) Computation latency is the time required for a task to be computed on the corresponding service device. This time is determined by the number of CPU cycles required for the task to be computed and the computing power of the service device itself. It is expressed as:
[0040] 3) Latency is the time from when a task is transmitted to the service device to when computation begins on that device. When the service device's computing power is F... s The maximum computing resources it possesses are D. s In this case, the task's waiting delay can be specifically defined as follows:
[0041] 4) Based on the four types of delays obtained previously, we can determine the total delay of a single task's complete non-local unloading process, expressed as: Similarly, the total latency of a single task's complete non-local unloading process can be expressed as: Therefore, in a system with a total of H tasks, we can determine the maximum tolerable latency T within the system. max The overall task latency ratio of the system is obtained as follows:
[0042] 5) For a given system with dimensions X... traffic ,Y traffic The study focuses on rectangular road networks, considering the maximum tolerable delay time of the task. The task is completed within the specified timeframe, and the uninstallation process is completed before the publisher leaves the road network, marked as "task successful". Otherwise, it is considered a "task failure". Therefore, we can calculate the task success rate within a single road network of the system by modeling it as follows: In vehicle-road cooperative systems, our goal is to minimize the average latency ratio while maximizing the task success rate.
[0043] 2. Based on the relative motion trajectories of vehicles and roadside units in the model, construct usable communication time links to determine stable communication conditions between vehicles and roadside units. This is done using any pair of intelligent connected vehicles (V-V) in the crossroads of the vehicle-road cooperative system. A With V B For example, the details are as follows:
[0044] 1) Based on the communication range and driving status of intelligent connected vehicles, define vehicle motion-related formulas, which can be specifically expressed through formulas. Calculate the available communication time for the vehicle to turn.
[0045] 2) Considering the intersection situation of vehicles, divided into straight-line travel and cross-traffic travel, the vehicle communication time is quantified, and the waiting delay caused by traffic lights is taken into account. Specifically, there are three categories: When car A and car B intersect, car A proceeds straight; when car A and car B intersect, car A merges into car B's lane traveling in the same direction; when car A and car B intersect, car A merges into car B's lane traveling in the opposite direction. Each category is then assigned a different... and The available communication time for the vehicle can be determined under all circumstances.
[0046] 3. The edge task offloading strategy of the vehicle-road cooperative system fusion feedback link perception as described in requirement 1 has the same characteristics as the common NP-hard problem "flexible job shop scheduling problem under uncertain working hours". This is a more complex combinatorial optimization problem than traditional flexible job shop scheduling. The optimization problem can be solved by the DGA algorithm, which combines Dijkstra's multi-policy generation algorithm and genetic algorithm, to determine the task offloading path and the overall task offloading strategy.
[0047] 1) Establish an intelligent connected vehicle communication network, generate the location of all devices in the road network at each time point, and store all information about the road network at each time point. Establish a task offloading strategy matrix M. i It is used to store all task unloading strategies and task unloading path plans in a single iteration simulation.
[0048] 2) Using Dijkstra's multi-strategy generation algorithm, the shortest current workshop distance is generated for all tasks. Based on the available communication time proposed in Requirement 2, the best unloading and transmission devices are allocated to the tasks to obtain the single-task unloading strategy matrix M in the current environment. x .
[0049] 3) Repeat the experiment multiple times to obtain the total task unloading strategy matrix N. j =[M1,…,M j ,…,M Y ], for each task, unload strategy matrix M i Construct a chromosome, treating individual task unloading strategies as genes, and perform crossover and mutation operations, iterating repeatedly until the optimal task unloading strategy matrix M for the system is obtained. y .
[0050] Example:
[0051] Now combined with the appendix Figure 1 The specific embodiments of the present invention are described as follows:
[0052] As attached Figure 1 As shown, the vehicle-to-infrastructure (V2I) system consists of three intelligent connected vehicles and four roadside units. Each roadside unit has an edge server. Three of these edge servers lack sufficient computing resources and can only perform task transmission rather than computation. The blue and red vehicles, each lacking sufficient computing resources, request a task. This task can be offloaded by an edge server with available computing resources or by the green vehicle. Different offloading strategies result in different task offloading paths and system efficiency.
[0053] First, we can construct a task offloading network based on the specific trajectories and communication status of intelligent connected vehicles and roadside units. Then, we can use Dijkstra's multi-strategy generation algorithm to obtain a large number of system task offloading strategies. Finally, we can use a genetic algorithm to continuously select, crossover, and mutate to obtain the optimal system task offloading strategy.
[0054] As can be seen from the above, the edge task offloading strategy proposed in this invention, which integrates feedback link perception in a vehicle-road cooperative system, demonstrates how technological innovation can improve task offloading efficiency and system utility in this field, providing a more efficient solution to the edge computing task offloading problem.
[0055] The above-described embodiments are intended to present a preferred implementation, but the scope of protection of this invention should not be limited thereto. Any equivalent substitutions or improvements made by those skilled in the art, based on the technical solutions and concepts disclosed in this invention, within the scope of this invention, should be considered to fall within the scope of protection of this invention's patent rights.
Claims
1. An edge task offloading strategy for a vehicle-road cooperative system that integrates feedback link perception, characterized in that... It includes the following steps: 1) The targeted vehicle-to-infrastructure (V2I) system comprises N roadside units and M intelligent IoT vehicles, denoted as device S = U∪V. Within the system, devices issue tasks, which can be transmitted or computed by any device in the system via V2I and V2V communication. Let device S be... i The r-th task issued is represented as l i,r The relevant attributes of its task are represented as follows These tasks will be offloaded to the appropriate devices via multi-hop transfer, provided that the devices support the types of computing and storage resources available, thereby reducing the load on the original devices. 2) The system design task unloading model is divided into local unloading and non-local unloading. Local unloading is divided into two parts: waiting latency and computation latency. Non-local unloading is divided into four parts: upload latency, waiting latency, computation latency, and return latency. The goal is to maximize system utility, that is, to minimize the average system latency while maximizing the task success rate. The problem is then optimized based on the task model and the multi-hop task unloading selection mechanism. 3) The problem is transformed into a "flexible workshop scheduling problem under uncertain working hours". The available communication time of the workshop is obtained by using vehicle movement trajectories and real-time road conditions. The DGA algorithm, which combines Dijkstra's multi-strategy generation algorithm and genetic algorithm, is used to solve the optimization problem and obtain the optimal task unloading strategy of the system.
2. The server resource reservation method for an edge computing system under fault scenarios as described in claim 1, characterized in that: The specific latency of a task is broken down into four main parts: upload latency, waiting latency, computation latency, and data return latency. In this process, unloading the task on the task publisher is called local unloading, while unloading on a non-task publisher is called non-local unloading. The local unloading process does not involve upload latency or data return latency. 1) For transmission latency, we divide it into upload latency and return latency. Upload latency represents the transmission time required for the task to be transmitted to the service device via the uplink, while return latency represents the time required for the task result to be transmitted back to the publishing device via the downlink. The transmission latency required for the same type of task will vary depending on the specific offloading link. Locally unloaded tasks are calculated on their own publisher device and do not require transmission or return, therefore they have no upload latency or return latency. Specifically, for tasks in non-local unloading processes, the upload latency is expressed as... The return latency is represented as 2) Computation latency is the time required for a task to be computed on the corresponding service device. This time is determined by the number of CPU cycles required for the task to be computed and the computing power of the service device itself. Specifically, it is expressed as: 3) Latency is the time from when a task is transmitted to the service device to when it begins computation on that device. Tasks within the device are queued according to an M / M / 1 queue model. The number of tasks a device in the system can execute within a given time period is determined by its maximum computing resource limit D. s When the computing power of the service device is F s In this case, the waiting latency of a task can be quantified, specifically as follows: 4) Based on the four types of delays obtained previously, we can determine the total delay of a single task's complete non-local unloading process, expressed as: Similarly, the total latency of a single task's complete non-local unloading process can be expressed as: Therefore, in a system with a total of H tasks, we can determine the maximum tolerable latency T within the system. max The overall task latency ratio of the system is obtained as follows: 5) To further evaluate the efficiency of system task unloading, we set up a system with dimensions X... traffic ,Y traffic The study focuses on rectangular road networks, considering the maximum tolerable delay time of the task. The task is completed within the specified timeframe, and the uninstallation process is marked as "task successful" before the publisher leaves the road network. Therefore, we can calculate the task success rate model within a single road network of the system as follows: In vehicle-road cooperative systems, our goal is to minimize the average latency ratio while maximizing the task success rate.
3. Based on the relative motion trajectories of vehicles and roadside units in the model, construct usable communication time links to determine stable communication conditions between vehicles and roadside units. This is done using any pair of intelligent connected vehicles (V-V) in the crossroads of the vehicle-road cooperative system. A With V B For example, the details are as follows: 1) Based on the communication range and driving status of intelligent connected vehicles, define vehicle motion-related formulas, which can be specifically expressed through formulas. Calculate the available communication time for the vehicle to turn. 2) Considering the intersection situation of vehicles, divided into straight-line travel and cross-traffic travel, the vehicle communication time is quantified, and the waiting delay caused by traffic lights is taken into account. Specifically, there are three categories: When car A and car B intersect, car A proceeds straight; when car A and car B intersect, car A merges into car B's lane traveling in the same direction; when car A and car B intersect, car A merges into car B's lane traveling in the opposite direction. Each category is then assigned a different... and The available communication time for the vehicle can be determined under all circumstances. 3) Based on the above analysis, the optimization problem jointly considering V2I and V2V in the vehicle-road cooperative system can be defined as: min αT average +(1-a)Q succ , (1) ∑ s∈S M ns ＝1, (2) ∑ s∈S M ms ＝1, (3) Among these constraints, constraint (2) ensures that when a roadside unit publishes a task, it can only offload it to one service device. Constraint (3) ensures that when a vehicle publishes a task, it can only offload it to one service device, thus preventing the task from being published repeatedly. Constraint (4) ensures that the device transmitting the task must have sufficient storage space to receive the corresponding task, preventing the receiving device from exceeding its capacity. Constraint (6) ensures that the task type corresponds to the task processing capacity of the device, and the publisher device must offload the generated task to a service device with the capability to calculate the corresponding task type.
4. The edge task offloading DGA strategy for the vehicle-road cooperative system fusion feedback link perception as described in claim 2, characterized by the following: Similar to the common NP-hard problem of "flexible job shop scheduling under uncertain working hours," this is a more complex combinatorial optimization problem than traditional flexible job shop scheduling. The optimization problem can be solved using the DGA algorithm, which combines Dijkstra's multi-policy generation algorithm and a genetic algorithm, to determine the task unloading path and the overall task unloading strategy. 1) Establish an intelligent connected vehicle communication network, generate the location of all devices in the road network at each time point, and store all information about the road network at each time point. Establish a task offloading strategy matrix M. i It is used to store all task unloading strategies and task unloading path plans in a single iteration simulation. 2) Using Dijkstra's multi-strategy generation algorithm, the shortest current workshop distance is generated for all tasks. Based on the available communication time proposed in Requirement 2, the best unloading and transmission devices are allocated to the tasks to obtain the single-task unloading strategy matrix M in the current environment. x . 3) Repeat the experiment multiple times to obtain the total task unloading strategy matrix N. j =[M1,…,M j ,…,M Y ], for each task, unload strategy matrix M i Construct a chromosome, treating individual task unloading strategies as genes, and perform crossover and mutation operations, iterating repeatedly until the optimal task unloading strategy matrix M for the system is obtained. y .