Low-delay vehicle active service caching and task unloading method for Internet of Vehicles

Abstract

The invention relates to a low-delay vehicle active service caching and task unloading method for the Internet of Vehicles. The method comprises the following steps: step 1, constructing a C-V2X-based vehicle edge computing system architecture; step 2, constructing a vehicle connection model based on social mobile perception, and establishing V2V pairing for task unloading and service acquisition; 3, constructing an edge-end cooperative vehicle task unloading model, and minimizing task completion time delay; 4, a multi-path collaborative vehicle active service cache model is constructed, and a service vehicle actively selects a service acquisition path with the minimum time delay; 5, establishing a series of constraint conditions according to the network topology, the time delay tolerance requirement and the resource limitation by taking the minimum total task completion time delay as a target function, and constructing an overall optimization problem; and step 6, solving to obtain an approximate optimal decision, and realizing vehicle active service caching and task unloading. According to the invention, the total vehicle task completion time delay is obviously reduced, and the system resource utilization rate is improved.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle network communication and edge computing technology, and specifically relates to a vehicle active service caching and task offloading method for low-latency vehicle network. Background Art

[0002] Vehicular edge computing (VEC), an emerging computing service paradigm in the connected vehicle (IoV), aims to provide low-latency services to nearby vehicles by empowering edge nodes with lightweight computing and storage resources. By deploying VEC servers on edge nodes, such as service vehicles and roadside units (RSUs), vehicles can offload pending tasks to edge nodes for processing via vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communications. Compared to traditional local computing and cloud computing solutions, the VEC architecture not only overcomes the limitations of the vehicle's own computing power but also significantly shortens task transmission distances, effectively addressing the surge in in-vehicle data traffic and the growing demand for low-latency services.

[0003] In a VEC network, task processing not only relies on the task data generated by the vehicle, but also requires edge nodes to cache related service content, such as the AI ​​models, databases, and executable software required for the task. This process is called edge service caching and is usually a prerequisite for edge nodes to complete task processing. However, due to the limited storage capacity of edge nodes, it is impossible to cache all service models. Once the service cache does not match the vehicle task requirements, it will significantly increase the task completion delay and even cause the task to timeout and fail. Therefore, facing the end-to-end low latency requirements of the Internet of Vehicles, the joint optimization of service caching, task offloading, and resource allocation has become a core challenge for resource-constrained VEC systems, and has received widespread attention from academia and industry at home and abroad.

[0004] Although existing research has explored the joint optimization of service caching, task offloading, and resource allocation in VEC networks, many shortcomings remain: most studies focus only on service caching strategies on the network side (such as RSUs), ignoring the potential and role of vehicle-side service caching. Unlike centralized network-side service caching, vehicle-side caching behavior is often associated with the driver's interests or task preferences. When a vehicle cache misses, the vehicle can leverage its mobility and autonomy to proactively obtain and cache the required services from the service provider. Therefore, the proactive service acquisition capability of the service vehicle becomes an effective supplement to the vehicle service cache. However, existing research rarely considers the vehicle service acquisition process. Even when it is involved, it is limited to downloading service models from RSUs and lacks systematic modeling and utilization of multi-source service acquisition mechanisms (such as V2V).

[0005] In practice, vehicles typically come pre-installed with service models related to their specific tasks and can share services with other vehicles through V2V communication. On the one hand, vehicle mobility and the low latency of V2V communication enhance the flexibility of service acquisition. On the other hand, vehicle-initiated service acquisition and task offloading face numerous challenges: 1) High-speed vehicle movement leads to rapid changes in network topology, which in turn causes intermittent connectivity issues. If the service vehicle is not properly selected, V2V transmission may be interrupted. 2) Vehicle-initiated service acquisition models introduce additional service transmission latency. Only when overall latency is optimized will the system tend to offload tasks to vehicles that have not yet cached services. 3) The complex coupling relationship between task offloading, service acquisition, and resource allocation increases the difficulty of joint optimization.

[0006] Therefore, it is urgent to design a method to jointly optimize vehicle active service acquisition, task offloading and resource allocation on the basis of ensuring the reliability of V2V connection, so as to further improve the low-latency service capability of the VEC network. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a low-latency vehicle-to-vehicle (V2V) proactive service caching and task offloading method. By constructing a vehicle pairing model based on social mobility awareness, the reliability and sustainability of vehicle-to-vehicle (V2V) connections are improved. Furthermore, the method jointly optimizes the vehicle's proactive service acquisition, task offloading decisions, and bandwidth and computing resource allocation. This allows vehicles to adaptively select service acquisition and task offloading methods based on network status, task characteristics, and resource constraints, significantly reducing overall vehicle task completion latency and improving system resource utilization.

[0008] Explanation of terms: 1. MATLAB intlinprog function is used to solve mixed integer linear programming. It mainly uses branch and bound method to obtain the optimal solution. The specific function call is: x=intlinprog(f,intcon,A,b,Aeq,beq,lb,ub); where x represents the decision variable matrix, f represents the objective function coefficient matrix corresponding to x, and intcon is a matrix from 1 to The integer sequence indicates that all variables in x are integers, A represents the coefficient matrix of the inequality constraint, b represents the right-hand side vector of the inequality constraint, Aeq represents the coefficient matrix of the equality constraint, beq represents the right-hand side vector of the equality constraint, lb is the lower bound vector of the decision variable x (all 0s), and ub is the upper bound vector of the decision variable x (all 1s).

[0009] 2. Interior point method is a type of algorithm used to solve linear and nonlinear convex optimization problems. It approaches the optimal solution by moving within the feasible region. It is an efficient polynomial time algorithm with polynomial time complexity.

[0010] The technical solution of the present invention is: A method for low-latency vehicle active service caching and task offloading for Internet of Vehicles (IoV) includes the following steps: Step 1: Build a C-V2X-based vehicle edge computing system architecture; Step 2: Combine vehicle connection duration and inter-vehicle social relationships to build a vehicle connection model based on social mobility awareness and establish V2V pairing for task offloading and service acquisition; Step 3: Based on the social mobility-aware vehicle connection model constructed in Step 2, a collaborative vehicle task offloading model is constructed to determine whether each vehicle's computing tasks should be processed locally, offloaded to a roadside unit (RSU), or offloaded to a paired service vehicle to minimize task completion latency. Step 4: Based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed, in which the serving vehicle actively selects the service acquisition path with the minimum latency. Step 5: Taking minimizing the total task completion delay as the objective function, a series of constraints are established based on the network topology, delay tolerance requirements, and resource limitations to construct an overall optimization problem. Step 6. Decompose the optimization problem constructed in step 5 into three sub-problems: task offloading, resource allocation, and service acquisition. Combined with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed. The approximate optimal decision is obtained through iterative solution to achieve vehicle-active service caching and task offloading.

[0011] According to the preferred embodiment of the present invention, a vehicle edge computing system architecture based on C-V2X is constructed; including: Considering the urban two-way lane vehicle networking scenario, the vehicle edge computing system architecture includes a roadside unit RSU and The set of all moving vehicles is ; Assume that the RSU coverage area is a radius of The RSU is circular and equipped with a VEC server that has cached all service models and provides computing services to vehicles in the RSU coverage area through V2I communication; The vehicle is equipped with a VEC server that caches part of the service model to handle its own tasks or tasks offloaded from the V2V link; Vehicles are divided into three categories based on whether they generate computing tasks and service willingness: The vehicle only generates computing tasks and does not provide any services; Vehicles not only have computing tasks, but can also provide computing services for other task-generating vehicles; The vehicle does not generate computing tasks, but only provides computing services; represents the set of mission vehicles, Represents a collection of service vehicles; make Indicates that the vehicle The generated tasks, where Indicates the data size, Indicates the amount of computing resources required, Indicates completion of the task Maximum tolerable delay; Let the service model set be represented as , where the size of each service model is , Indicates processing tasks Service Model .

[0012] According to a preferred embodiment of the present invention, a vehicle connection model based on social mobility perception is constructed by combining vehicle connection duration and inter-vehicle social relationships; the model includes: Step 2.1, order The coordinates of the vehicle are ,in represents the vehicle antenna height; the RSU coordinates are ,in Indicates the RSU antenna height; Step 2.2: Use formula (1) to construct the vehicle Duration of V2I communication connection with RSU: ; In formula (1), Indicates vehicle The speed of the vehicle is positive when it moves from left to right and negative when it moves from right to left. Indicates absolute value; Step 2.3: Use formula (2) to construct the vehicle With vehicle Duration of V2V communication connection: ; In formula (2), L represents the maximum V2V communication distance; Step 2.4: Use formula (3) to construct the vehicle With vehicle V2V physical connection indicators between: ; In formula (3), Indicates the V2V connection duration threshold; Step 2.5: Use the social graph Modeling social relationships between vehicle users, where Represents the set of social relationships between vehicle users; Represents the normalized weighted values ​​of multiple social attributes, including historical connection records, trust, reputation, destination similarity, and service preferences; Step 2.6: Combine the V2V physical connection index and the social relationship between vehicle users and use formula (4) to establish the vehicle V2V pairing index based on social mobility perception, that is, the vehicle connection model based on social mobility perception: ; In formula (4), Mission Vehicle A collection of service vehicles that are physically connected to each other via V2V.

[0013] Preferably, according to the present invention, based on the social mobility-aware vehicle connection model constructed in step 2, a vehicle task offloading model with edge collaboration is constructed, including: Step 3.1. Definition represents the vehicle task unloading decision matrix, where represents the task offloading decision, , j=i represents vehicle Processing tasks locally ; , j≠i represents vehicle The task Offloading to edge nodes to process; Step 3.2: , when j≠i, use formula (5) to express the task Transmission delay of tasks offloaded to edge nodes: ; in, Indicates the allocation to the vehicle Uplink bandwidth, Indicates vehicle The transmission power, represents the noise power, represents the average interference caused by spectrum reuse by non-adjacent RSUs, Indicates vehicle To the edge node The uplink block fading channel coefficient is represents the path loss, represents the path loss exponent; Step 3.3: Use formula (6) to express the task The task calculation delay is: ; in, , when j=i, it means vehicle For the task allocated local computing resources; , when j≠i, it represents an edge node For the task Allocated computing resources.

[0014] Preferably, according to the present invention, based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed; including: Step 4.1: Assume that the initial cache model set of the service vehicle is ,in Representation service model Cached in service vehicle otherwise 0; Step 4.2: and If the service vehicle There are still tasks to be processed If there is a willingness, the service model is obtained from the service provider. ; Let the set of service providers be represented as ,definition represents the vehicle service acquisition decision matrix, where Indicates the service acquisition decision. If the service vehicle From the service provider Get the service model ,but ;otherwise, .

[0015] Step 4.3: , when i=0, service vehicle Choose to download the service model from RSU and use formula (7) to express the service vehicle The latency of receiving services from RSU: ; in, Indicates the transmit power of RSU, Indicates the transfer from RSU to service vehicle The downlink block fading channel coefficient is Indicates the transfer from RSU to service vehicle The downlink path loss is Indicates the allocation to the vehicle The downlink bandwidth of the service vehicle The task completion delay of downloading the service model from RSU is expressed as formula (8): ; Step 4.4: , When the service vehicle Choose from mission vehicles Upload the service model and use formula (9) to express the service vehicle From the mission vehicle Service acquisition delay at: ; service vehicles Choose from mission vehicles The task completion delay of the upload service model can be expressed as formula (10): ; Step 4.5: Combining the two service acquisition modes of steps 4.4 and 4.5, we use formula (11) to establish a vehicle active service cache model based on the task completion delay of vehicle active service acquisition, i.e., multi-path collaboration: .

[0016] Preferably, according to the present invention, minimizing the total task completion delay is used as the objective function, a series of constraints are established according to the network topology, delay tolerance requirements and resource constraints, and an overall optimization problem is constructed; including: Step 5.1: Use formula (12) to construct the objective function that minimizes the total task completion delay: ; The optimization variables are: represents the vehicle task unloading decision matrix, represents the computing resource allocation decision matrix, represents the bandwidth resource allocation decision matrix, represents the vehicle service acquisition decision matrix; Step 5.2: Use formula (13) to construct the constraint conditions: ; in, Indicates that each service vehicle can only select one service provider to obtain the service model; Indicates that the service vehicle's service acquisition delay from the RSU cannot exceed the duration of the V2I connection between the vehicle and the RSU; Indicates that the total amount of service vehicle cache and retrieval models cannot exceed its maximum storage capacity ; Indicates that all tasks are inseparable and each task must and can only be processed on one computing node; Indicates the local computation latency of the task The delay tolerance requirement of the task cannot be exceeded; The completion delay of the task offloaded to the RSU must not only meet the task delay tolerance requirement, but also cannot exceed the V2I connection duration. Indicates vehicle The task Whether to offload the decision variables to RSU processing, Indicates a task The transmission delay of the task offloaded to the RSU, Indicates a task Task calculation delay at the RSU; The completion delay of the task offloading to the service vehicle must not only meet the task delay tolerance requirement, but also cannot exceed the V2V connection duration; Represents a compute node For the task The allocated computing resources cannot exceed its total computing capacity ; Represented as a task The allocated uplink bandwidth resources cannot exceed the total system bandwidth ; Indicates the value constraints of optimization variables; Preferably, according to the present invention, the optimization problem constructed in step 5 is decomposed into three sub-problems: task offloading, resource allocation, and service acquisition. In combination with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed to obtain an approximately optimal decision through iterative solution. This includes: Step 6.1: Fix the optimization variables , use formula (14) to construct the task offloading sub-problem: ; Step 6.2: Fix the optimization variables , use formula (15) to construct the bandwidth allocation and computing resource allocation sub-problems: ; Step 6.3: Fix the optimization variables , using formula (16) to construct the service acquisition sub-problem: ; Step 6.4: Design a heuristic algorithm with polynomial time complexity for the service acquisition subproblem constructed in Step 6.3 to obtain a vehicle service acquisition decision with near-optimal performance. Step 6.5: Construct an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition to obtain an approximately optimal decision through iterative solution.

[0017] Further preferably, the heuristic algorithm of polynomial time complexity includes: Step 6.4.1. For each service vehicle and each mission vehicle , calculate the task offloading delay according to formula (5) in step 3.2 ; Step 6.4.2: For each service model ,like , then the corresponding service acquisition decision is set to zero, , Otherwise, calculate the service acquisition delay from the RSU according to steps 4.3 and 4.4 respectively and from vehicles Service acquisition latency ; Step 6.4.3, Execution judgment: When and When Otherwise, set ; Step 6.4.4, determine the constraints: If the constraints are met at this time , then the loop is exited and the final vehicle service acquisition decision with service acquisition close to the optimal performance is output. ; Otherwise, set the service acquisition decision to 0 and continue the loop.

[0018] Further preferably, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed to iteratively solve and obtain a near-optimal decision, including: Step 6.5.1. Initialization: Setting the initial task offloading decision matrix , initial computing resource allocation decision matrix , initial bandwidth allocation decision matrix , Initial vehicle service acquisition decision matrix , set the number of iteration steps , and include the parameters Enter the VEC server; Step 6.5.2, variable calculation: fixed , They represent the computing resource allocation decision matrix, bandwidth allocation decision matrix, and vehicle service acquisition decision matrix at the rth iteration step, respectively. The MATLAB intlinprog function is used to solve the task offloading subproblem constructed in step 6.1, and we get ; represents the updated task offloading decision matrix at the current iteration r+1 steps; fixed , use the interior point method to solve the bandwidth allocation and computing resource allocation sub-problems constructed in step 6.2, and get ; Represents the updated computing resource allocation decision matrix under the current iteration r+1 steps; represents the updated bandwidth allocation decision matrix at the current iteration r+1 steps; fixed , use the heuristic algorithm with polynomial time complexity designed in step 6.4 to solve the service acquisition subproblem constructed in step 6.3, and get ; Represents the updated service acquisition decision matrix at the current iteration r+1 steps; Step 6.5.3, Determination: Determine the value of the objective function Is it equal to If they are equal, the loop is terminated and the variable matrix is ​​output. ; If not equal, proceed to step 6.5.4; Step 6.5.4, variable update: Updated to ,make , and repeat step 6.5.2; The final approximate optimal decision is a set of matrices obtained when the iterative algorithm converges ,in represents the optimal task offloading decision matrix, represents the optimal computing resource allocation decision matrix, represents the optimal bandwidth allocation decision matrix, Represents the optimal service acquisition decision matrix.

[0019] Further preferably, the vehicle proactive service caching and task offloading are implemented according to the obtained approximate optimal decision, including: Task offloading: Based on , each mission vehicle Know your mission Where to uninstall: If , the task is used locally computing resources for processing; if , then the mission vehicle uses Bandwidth offloads tasks to RSU, RSU uses The computing resources for task processing; if a service vehicle , , then the mission vehicle uses Bandwidth offloads tasks to service vehicles , service vehicles use computing resources for task processing; Active service acquisition: When the task Unloaded onto service vehicle ( ) and service vehicles No caching required for serving models When the service vehicle will be based on Active service acquisition process: If , then the service vehicle Actively obtain service models from RSU ;like , then the service vehicle Active slave mission vehicle Get the service model .

[0020] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned vehicle active service caching and task offloading method for low latency in the Internet of Vehicles.

[0021] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned vehicle active service caching and task offloading method for low-latency vehicle networking.

[0022] The beneficial effects of the present invention are: This paper proposes a low-latency vehicle-to-vehicle (V2V) proactive service caching and task offloading method. By constructing a vehicle pairing model based on social mobility awareness, the method improves the reliability and sustainability of vehicle-to-vehicle (V2V) connections. Based on this, the method jointly optimizes the vehicle's proactive service acquisition, task offloading decisions, and bandwidth and computing resource allocation. A low-complexity iterative algorithm based on block coordinate descent is designed to rapidly achieve near-optimal decisions. This enables vehicles to adaptively select service acquisition and task offloading methods across multiple paths based on network status, task characteristics, and resource constraints, significantly reducing overall vehicle task completion latency and improving system resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a diagram of the vehicle edge computing system architecture to which the present invention is applicable; Figure 2This is a flow chart of the iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition proposed in the present invention; Figure 3 This is a comparison chart of the total task completion delay between the method proposed in the present invention and the existing method under different numbers of vehicles. DETAILED DESCRIPTION

[0024] The present invention will be further defined below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0025] Example 1 A method for low-latency vehicle active service caching and task offloading for Internet of Vehicles (IoV) includes the following steps: Step 1: Build a C-V2X-based vehicle edge computing system architecture; Step 2: Combine the vehicle connection duration and the social relationship between vehicles to build a vehicle connection model based on social mobility perception to establish reliable and continuous V2V pairing for task offloading and service acquisition; that is, determine .

[0026] Step 3: Based on the social mobility-aware vehicle connection model constructed in Step 2, a collaborative vehicle task offloading model is constructed to determine whether each vehicle's computing tasks should be processed locally, offloaded to a roadside unit (RSU), or offloaded to a paired service vehicle to minimize task completion latency. Step 4: Based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed. This multi-path coordinated vehicle active service caching model addresses the situation where the serving vehicle selected as the task offloading target does not cache the required service model. The serving vehicle actively selects the service acquisition path with the shortest latency. Step 5: Taking minimizing the total task completion delay as the objective function, a series of constraints are established based on the network topology, delay tolerance requirements, and resource limitations to construct an overall optimization problem. Step 6. Decompose the optimization problem constructed in step 5 into three sub-problems: task offloading, resource allocation, and service acquisition. Combined with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed. The approximate optimal decision is obtained through iterative solution to achieve vehicle-active service caching and task offloading.

[0027] Example 2 The difference between the method for low-latency vehicle active service caching and task offloading for the Internet of Vehicles described in Example 1 is that: Build a vehicle edge computing system architecture based on C-V2X; Figure 1 Shown; including: Considering the urban two-way lane vehicle networking scenario, the vehicle edge computing system architecture includes a roadside unit RSU and The set of all moving vehicles is ; Assume that the RSU coverage area is a radius of The RSU is circular and equipped with a VEC server that has cached all service models, and provides computing services to vehicles in the RSU coverage area through V2I communication; in the present invention, "service model" refers to the software components, trained AI models or related databases necessary to perform computing tasks.

[0028] The vehicle is equipped with a VEC server that caches part of the service model to handle its own tasks or tasks offloaded from the V2V link; Vehicles are divided into three categories based on whether they generate computing tasks and service willingness: The vehicle only generates computing tasks and does not provide any services; Vehicles not only have computing tasks, but can also provide computing services for other task-generating vehicles; The vehicle does not generate computing tasks, but only provides computing services; represents the set of mission vehicles, Represents a collection of service vehicles; make Indicates that the vehicle The generated tasks, where Indicates the data size, Indicates the amount of computing resources required, Indicates completion of the task Maximum tolerable delay; Let the service model set be represented as , where the size of each service model is , Indicates processing tasks Service Model .

[0029] Combining vehicle connection duration with inter-vehicle social relationships, a vehicle connection model based on social mobility awareness is constructed; this includes: Step 2.1, order The coordinates of the vehicle are ,in represents the vehicle antenna height; the RSU coordinates are ,in Indicates the RSU antenna height; Step 2.2: Use formula (1) to construct the vehicle Duration of V2I communication connection with RSU: ; In formula (1), Indicates vehicle The speed of the vehicle is positive when it moves from left to right and negative when it moves from right to left. Indicates absolute value; Step 2.3: Use formula (2) to construct the vehicle With vehicle Duration of V2V communication connection: ; In formula (2), L represents the maximum V2V communication distance; Step 2.4: Use formula (3) to construct the vehicle With vehicle V2V physical connection indicators between: ; In formula (3), Indicates the V2V connection duration threshold; Step 2.5: Use the social graph Modeling social relationships between vehicle users, where Represents the set of social relationships between vehicle users; Represents the normalized weighted values ​​of multiple social attributes, including historical connection history, trust, reputation, destination similarity, and service preferences. Social attributes can be effectively extracted from relevant data sources (such as mapping software, navigation software, social networking applications, and in-vehicle service platforms) using technologies such as deep learning. For example, historical connection history is a count of connection frequency; trust and reputation are derived from rating systems or interaction histories within in-vehicle service platforms; destination similarity can be calculated from navigation data; and service preferences can be extracted from social networking applications or explicitly set by the user.

[0030] Step 2.6: Combine the V2V physical connection index and the social relationship between vehicle users and use formula (4) to establish the vehicle V2V pairing index based on social mobility perception, that is, the vehicle connection model based on social mobility perception: ; In formula (4), Mission Vehicle A collection of service vehicles that are physically connected to each other via V2V.

[0031] Based on the social mobility-aware vehicle connection model constructed in step 2, a vehicle task offloading model with edge collaboration is constructed, including: Step 3.1. Definition represents the vehicle task unloading decision matrix, where represents the task offloading decision, , j=i represents vehicle Processing tasks locally ; , j≠i represents vehicle The task Offloading to edge nodes Otherwise it is 0.

[0032] Step 3.2: , when j≠i, use formula (5) to express the task The transmission delay of tasks offloaded to edge nodes (RSU or service vehicles): ; in, Indicates the allocation to the vehicle Uplink bandwidth, Indicates vehicle The transmission power, represents the noise power, represents the average interference caused by spectrum reuse by non-adjacent RSUs, Indicates vehicle To the edge node The uplink block fading channel coefficient is represents the path loss, represents the path loss exponent; Step 3.3: Use formula (6) to express the task The task calculation delay is: ; in, , when j=i, it means vehicle For the task allocated local computing resources; , when j≠i, it represents an edge node For the task Allocated computing resources.

[0033] The constructed edge-end collaborative vehicle task offloading model mainly includes the task offloading decision matrix And the formulas for calculating the related delay (Formula (5) and Formula (6)).

[0034] Based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed; including: Step 4.1: Assume that the initial cache model set of the service vehicle is ,in Representation service model Cached in service vehicle otherwise 0; Step 4.2: and If the service vehicle There are still tasks to be processed If there is a willingness, the service model is obtained from the service provider (RSU or other vehicles) ; Let the set of service providers be represented as ,definition represents the vehicle service acquisition decision matrix, where Indicates the service acquisition decision. If the service vehicle From the service provider Get the service model ,but ;otherwise, .

[0035] Step 4.3: , when i=0, service vehicle Choose to download the service model from RSU and use formula (7) to express the service vehicle Delay obtained from the service at RSU: ; in, Indicates the transmit power of RSU, Indicates the transfer from RSU to service vehicle The downlink block fading channel coefficient is Indicates the transfer from RSU to service vehicle The downlink path loss is Indicates the allocation to the vehicle Downlink bandwidth; Since uplink task offloading and downlink service acquisition are parallel transmission processes, the service vehicle The task completion delay of downloading the service model from RSU is expressed as formula (8): ; Step 4.4: , When the service vehicle Choose from mission vehicles Upload the service model and use formula (9) to express the service vehicle From the mission vehicle Service acquisition delay at: ; Since uplink task offloading and uplink service acquisition are serial transmission processes, the service vehicle Choose from mission vehicles The task completion delay of the upload service model can be expressed as formula (10): ; Step 4.5: Combining the two service acquisition modes of steps 4.4 and 4.5, we use formula (11) to establish a vehicle active service cache model based on the task completion delay of vehicle active service acquisition, i.e., multi-path collaboration: .

[0036] Taking minimizing the total task completion delay as the objective function, a series of constraints are established based on the network topology, delay tolerance requirements, and resource limitations to construct an overall optimization problem; this includes: Step 5.1: Use formula (12) to construct the objective function that minimizes the total task completion delay: ; The optimization variables are: represents the vehicle task unloading decision matrix, represents the computing resource allocation decision matrix, represents the bandwidth resource allocation decision matrix, represents the vehicle service acquisition decision matrix; Step 5.2: Use formula (13) to construct the constraint conditions: ; in, Indicates that each service vehicle can only select one service provider to obtain the service model; Indicates that the service vehicle's service acquisition delay from the RSU cannot exceed the duration of the V2I connection between the vehicle and the RSU; Indicates that the total amount of service vehicle cache and retrieval models cannot exceed its maximum storage capacity ; Indicates that all tasks are inseparable and each task must and can only be processed on one computing node; Indicates the local computation latency of the task The delay tolerance requirement of the task cannot be exceeded; The completion delay of the task offloaded to the RSU must not only meet the task delay tolerance requirement, but also cannot exceed the V2I connection duration. Indicates vehicle The task Whether to offload the decision variable to RSU processing, Indicates a task The transmission delay of the task offloaded to the RSU, Indicates a task Task calculation delay at the RSU; The completion delay of the task offloading to the service vehicle must not only meet the task delay tolerance requirement, but also cannot exceed the V2V connection duration; Represents a compute node For the task The allocated computing resources cannot exceed its total computing capacity ; Represented as a task The allocated uplink bandwidth resources cannot exceed the total system bandwidth ; Indicates the value constraints of optimization variables; The optimization problem constructed in step 5 is decomposed into three sub-problems: task offloading, resource allocation, and service acquisition. In combination with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed to iteratively solve and obtain a near-optimal decision. This includes: Step 6.1: Fix the optimization variables , use formula (14) to construct the task offloading sub-problem: ; Step 6.2: Fix the optimization variables , use formula (15) to construct the bandwidth allocation and computing resource allocation sub-problems: ; Step 6.3: Fix the optimization variables , using formula (16) to construct the service acquisition sub-problem: ; Step 6.4: Design a heuristic algorithm with polynomial time complexity for the service acquisition subproblem constructed in Step 6.3, which can quickly obtain a vehicle service acquisition decision with near-optimal performance. Step 6.5: Construct an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition to obtain an approximately optimal decision through iterative solution.

[0037] Heuristic algorithms with polynomial time complexity include: Step 6.4.1. For each service vehicle and each mission vehicle , calculate the task offloading delay according to formula (5) in step 3.2 ; Step 6.4.2: For each service model ,like , then the corresponding service acquisition decision is set to zero, , Otherwise, calculate the service acquisition delay from the RSU according to steps 4.3 and 4.4 respectively and from vehicles Service acquisition latency ; Step 6.4.3, Execution judgment: When and When Otherwise, set ; Step 6.4.4, determine the constraints: If the constraints are met at this time , then jump out of the loop and output the final vehicle service acquisition decision with service acquisition close to the optimal performance ; Otherwise, set the service acquisition decision to 0 and continue the loop.

[0038] Construct an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition to obtain approximate optimal decisions through iterative solutions. The algorithm flow chart is as follows: Figure 2 Shown: Includes: Step 6.5.1. Initialization: Setting the initial task offloading decision matrix , initial computing resource allocation decision matrix , initial bandwidth allocation decision matrix , Initial vehicle service acquisition decision matrix , set the number of iteration steps , and include the parameters Enter the VEC server; Step 6.5.2, variable calculation: fixed , They represent the computing resource allocation decision matrix, bandwidth allocation decision matrix, and vehicle service acquisition decision matrix at the rth iteration step, respectively. The MATLAB intlinprog function is used to solve the task offloading subproblem constructed in step 6.1, and we get ; represents the updated task offloading decision matrix at the current iteration r+1 steps; fixed , Represents the service acquisition decision matrix at the rth iteration step; use the interior point method to solve the bandwidth allocation and computing resource allocation sub-problems constructed in step 6.2, and obtain ; Represents the updated computing resource allocation decision matrix under the current iteration r+1 steps; represents the updated bandwidth allocation decision matrix at the current iteration r+1 steps; fixed , use the heuristic algorithm with polynomial time complexity designed in step 6.4 to solve the service acquisition subproblem constructed in step 6.3, and get ; represents the task offloading decision matrix at the rth iteration step; Represents the computing resource allocation decision matrix at the rth iteration step; represents the bandwidth allocation decision matrix at the rth iteration step; Represents the updated service acquisition decision matrix at the current iteration r+1 steps; Step 6.5.3, Determination: Determine the value of the objective function Is it equal to , if they are equal, terminate the loop and output the variable matrix ; If not equal, proceed to step 6.5.4; Step 6.5.4, variable update: Updated to ,make , and repeat step 6.5.2; The final approximate optimal decision is a set of matrices obtained when the iterative algorithm (step 6) converges ,in represents the optimal task offloading decision matrix, represents the optimal computing resource allocation decision matrix, represents the optimal bandwidth allocation decision matrix, Represents the optimal service acquisition decision matrix.

[0039] Based on the obtained approximate optimal decision, the vehicle proactively caches services and offloads tasks, including: Task offloading: Based on , each mission vehicle Know your mission Where to uninstall: If , the task is used locally computing resources for processing; if , then the mission vehicle uses Bandwidth offloads tasks to RSU, RSU uses The computing resources for task processing; if a service vehicle , , then the mission vehicle uses Bandwidth offloads tasks to service vehicles , service vehicles use computing resources for task processing; Active service acquisition: When the task Unloaded onto service vehicle ( ) and service vehicles No caching required for serving models When the service vehicle will be based on Active service acquisition process: If , then the service vehicle Actively obtain service models from RSU ;like , then the service vehicle Active slave mission vehicle Get the service model This retrieved service model is then used and cached.

[0040] By following the best decision , achieving efficient task offloading, and when the task is offloaded to the service vehicle and the service vehicle has a cache miss, the service vehicle will actively obtain and cache the required service model, significantly reducing the total task completion delay.

[0041] Figure 3 A comparison chart of the total task completion delay of the invented method and the existing method under different numbers of vehicles is shown. Among them, comparison method 1 is a purely local calculation method, comparison method 2 is a calculation method with only RSU assistance, comparison method 3 is a no-service acquisition method, comparison method 4 is a fixed service acquisition method, and comparison method 5 is a resource average allocation method. It can be seen that the method of the present invention can achieve the lowest task completion delay. Specifically, the average task completion delay of the method of the present invention is reduced by 53.59%-77.65% compared with comparison method 1, 0.2%-35.42% compared with comparison method 2, 0.2%-26.68% compared with comparison method 3, 0.2%-13.74% compared with comparison method 4, and 9.38%-30.68% compared with comparison method 5.

[0042] The method of the present invention can still maintain the optimal task completion delay performance when the number of vehicles increases. ), since each vehicle can obtain sufficient bandwidth and computing resources, the invented method has only a small difference in total task completion latency compared with other comparison methods; when the number of vehicles increases, the number of task requests increases accordingly, and the bandwidth and computing resources allocated to each vehicle decrease, resulting in an increase in the overall task completion latency; at this time, the method of the present invention can significantly outperform other comparison methods by flexibly scheduling tasks and service models within the system and efficiently utilizing all available resources, thereby demonstrating its scheduling optimization capabilities in high-load scenarios.

[0043] Example 3 A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the vehicle active service caching and task offloading method for low-latency vehicle networking described in Example 1 or 2 are implemented.

[0044] Example 4 A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the vehicle active service caching and task offloading method for low-latency vehicle networking described in Example 1 or 2.

Claims

1. A low-latency vehicle active service caching and task offloading method for Internet of Vehicles, characterized by: The following steps are involved: Step 1: Build a C-V2X-based vehicle edge computing system architecture; Step 2: Combine vehicle connection duration and inter-vehicle social relationships to build a vehicle connection model based on social mobility awareness and establish V2V pairing for task offloading and service acquisition; Step 3: Based on the social mobility-aware vehicle connection model constructed in Step 2, a collaborative vehicle task offloading model is constructed to determine whether each vehicle's computing tasks should be processed locally, offloaded to a roadside unit (RSU), or offloaded to a paired service vehicle to minimize task completion latency. Step 4: Based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed, in which the serving vehicle actively selects the service acquisition path with the minimum latency. Step 5: Taking minimizing the total task completion delay as the objective function, a series of constraints are established based on the network topology, delay tolerance requirements, and resource limitations to construct an overall optimization problem. Step 6. Decompose the optimization problem constructed in step 5 into three sub-problems: task offloading, resource allocation, and service acquisition. Combined with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed. The approximate optimal decision is obtained through iterative solution to achieve vehicle-active service caching and task offloading.

2. A low-latency vehicle active service caching and task offloading method for Internet of Vehicles according to claim 1, characterized in that: Build a C-V2X-based vehicle edge computing system architecture; including: Considering the urban two-way lane vehicle networking scenario, the vehicle edge computing system architecture includes a roadside unit RSU and The set of all moving vehicles is ; Assume that the RSU coverage area is a radius of The RSU is circular and equipped with a VEC server that has cached all service models and provides computing services to vehicles in the RSU coverage area through V2I communication; The vehicle is equipped with a VEC server that caches part of the service model to handle its own tasks or tasks offloaded from the V2V link; Vehicles are divided into three categories based on whether they generate computing tasks and service willingness: The vehicle only generates computing tasks and does not provide any services; Vehicles not only have computing tasks, but can also provide computing services for other task-generating vehicles; The vehicle does not generate computing tasks, but only provides computing services; represents the set of mission vehicles, Represents a collection of service vehicles; make Indicates that the vehicle The generated tasks, where Indicates the data size, Indicates the amount of computing resources required, Indicates completion of the task Maximum tolerable delay; Let the service model set be represented as , where the size of each service model is , Indicates processing tasks Service Model .

3. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 1, characterized in that: Combining vehicle connection duration with inter-vehicle social relationships, a vehicle connection model based on social mobility awareness is constructed; this includes: Step 2.1, order The coordinates of the vehicle are ,in represents the vehicle antenna height; the RSU coordinates are ,in Indicates the RSU antenna height; Step 2.2: Use formula (1) to construct the vehicle Duration of V2I communication connection with RSU: ； In formula (1), Indicates vehicle The speed of the vehicle is positive when it moves from left to right and negative when it moves from right to left. Indicates absolute value; Step 2.3: Use formula (2) to construct the vehicle With vehicle Duration of V2V communication connection: ； In formula (2), L represents the maximum V2V communication distance; Step 2.4: Use formula (3) to construct the vehicle With vehicle V2V physical connection indicators between: ； In formula (3), Indicates the V2V connection duration threshold; Step 2.5: Use the social graph Modeling social relationships between vehicle users, where Represents the set of social relationships between vehicle users; Represents the normalized weighted values ​​of multiple social attributes, including historical connection records, trust, reputation, destination similarity, and service preferences; Step 2.6: Combine the V2V physical connection index and the social relationship between vehicle users and use formula (4) to establish the vehicle V2V pairing index based on social mobility perception, that is, the vehicle connection model based on social mobility perception: ； In formula (4), Mission Vehicle A collection of service vehicles that are physically connected to each other via V2V.

4. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 1, characterized in that: Based on the social mobility-aware vehicle connection model constructed in step 2, a vehicle task offloading model with edge collaboration is constructed, including: Step 3.

1. Definition represents the vehicle task unloading decision matrix, where represents the task offloading decision, , j=i represents vehicle Processing tasks locally ; , j≠i represents vehicle The task Offloading to edge nodes to process; Step 3.2: , when j≠i, use formula (5) to express the task Transmission delay of tasks offloaded to edge nodes: ； in, Indicates the allocation to the vehicle Uplink bandwidth, Indicates vehicle The transmission power, represents the noise power, represents the average interference caused by spectrum reuse by non-adjacent RSUs, Indicates vehicle To the edge node The uplink block fading channel coefficient is represents the path loss, represents the path loss exponent; Step 3.3: Use formula (6) to express the task The task calculation delay is: ； in, , when j=i, it means vehicle For the task allocated local computing resources; , when j≠i, it represents an edge node For the task Allocated computing resources.

5. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 1, characterized in that: Based on the edge-coordinated vehicle task offloading model constructed in step 3, a multi-path coordinated vehicle active service caching model is constructed; including: Step 4.1: Assume that the initial cache model set of the service vehicle is ,in Representation service model Cached in service vehicle otherwise 0; Step 4.2: and If the service vehicle There are still tasks to be processed If there is a willingness, the service model is obtained from the service provider. ; Let the set of service providers be represented as ,definition represents the vehicle service acquisition decision matrix, where Indicates the service acquisition decision. If the service vehicle From the service provider Get the service model ,but ;otherwise, ; Step 4.3: , when i=0, service vehicle Choose to download the service model from RSU and use formula (7) to express the service vehicle Delay obtained from the service at RSU: ； in, Indicates the transmit power of RSU, Indicates the transfer from RSU to service vehicle The downlink block fading channel coefficient is Indicates the transfer from RSU to service vehicle The downlink path loss is Indicates the allocation to the vehicle The downlink bandwidth of the service vehicle The task completion delay of downloading the service model from RSU is expressed as formula (8): ； Step 4.4: , When the service vehicle Choose from mission vehicles Upload the service model and use formula (9) to express the service vehicle From the mission vehicle Service acquisition delay at: ； service vehicles Choose from mission vehicles The task completion delay of the upload service model can be expressed as formula (10): ； Step 4.5: Combining the two service acquisition modes of steps 4.4 and 4.5, we use formula (11) to establish a vehicle active service cache model based on the task completion delay of vehicle active service acquisition, i.e., multi-path collaboration: 。 6. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 1, characterized in that: Taking minimizing the total task completion delay as the objective function, a series of constraints are established based on the network topology, delay tolerance requirements, and resource limitations to construct an overall optimization problem; this includes: Step 5.1: Use formula (12) to construct the objective function that minimizes the total task completion delay: ； The optimization variables are: represents the vehicle task unloading decision matrix, represents the computing resource allocation decision matrix, represents the bandwidth resource allocation decision matrix, represents the vehicle service acquisition decision matrix; Step 5.2: Use formula (13) to construct the constraint conditions: ； in, Indicates that each service vehicle can only select one service provider to obtain the service model; Indicates that the service vehicle's service acquisition delay from the RSU cannot exceed the duration of the V2I connection between the vehicle and the RSU; Indicates that the total amount of service vehicle cache and retrieval models cannot exceed its maximum storage capacity ; Indicates that all tasks are inseparable and each task must and can only be processed on one computing node; Indicates the local computation latency of the task The delay tolerance requirement of the task cannot be exceeded; The completion delay of the task offloaded to the RSU must not only meet the task delay tolerance requirement, but also cannot exceed the V2I connection duration. Indicates vehicle The task Whether to offload the decision variable to RSU processing, Indicates a task The transmission delay of the task offloaded to the RSU, Indicates a task Task calculation delay at the RSU; The completion delay of the task offloaded to the service vehicle must not only meet the task delay tolerance requirement, but also cannot exceed the V2V connection duration; Represents a compute node For the task The allocated computing resources cannot exceed its total computing capacity ; Represented as a task The allocated uplink bandwidth resources cannot exceed the total system bandwidth ; Represents the value constraints of optimization variables.

7. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 1, characterized in that: The optimization problem constructed in step 5 is decomposed into three sub-problems: task offloading, resource allocation, and service acquisition. In combination with the block coordinate descent method, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed to obtain an approximately optimal decision through iterative solution. include: Step 6.1: Fix the optimization variables , use formula (14) to construct the task offloading sub-problem: ； Step 6.2: Fix the optimization variables , use formula (15) to construct the bandwidth allocation and computing resource allocation sub-problems: ； Step 6.3: Fix the optimization variables , using formula (16) to construct the service acquisition sub-problem: ； Step 6.4: Design a heuristic algorithm with polynomial time complexity for the service acquisition subproblem constructed in Step 6.3 to obtain a vehicle service acquisition decision with near-optimal performance. Step 6.5: Construct an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition, and iteratively solve to obtain a near-optimal decision. Further preferably, the heuristic algorithm with polynomial time complexity includes: Step 6.4.

1. For each service vehicle and each mission vehicle , calculate the task offloading delay according to formula (5) in step 3.2 ; Step 6.4.2: For each service model ,like , then the corresponding service acquisition decision is set to zero, , Otherwise, calculate the service acquisition delay from the RSU according to steps 4.3 and 4.4 respectively and from vehicles Service acquisition latency ; Step 6.4.3, Execution judgment: When and When Otherwise, set ; Step 6.4.4, determine the constraints: If the constraints are met at this time , then jump out of the loop and output the final vehicle service acquisition decision with service acquisition close to the optimal performance ; Otherwise, set the service acquisition decision to 0 and continue the loop; Further preferably, an iterative optimization algorithm for joint task offloading, bandwidth allocation, computing resource allocation, and service acquisition is constructed to iteratively solve and obtain a near-optimal decision, including: Step 6.5.

1. Initialization: Setting the initial task offloading decision matrix , initial computing resource allocation decision matrix , initial bandwidth allocation decision matrix , Initial vehicle service acquisition decision matrix , set the number of iteration steps , and include the parameters Enter the VEC server; Step 6.5.2, variable calculation: fixed , They represent the computing resource allocation decision matrix, bandwidth allocation decision matrix, and vehicle service acquisition decision matrix at the rth iteration step, respectively. The MATLAB intlinprog function is used to solve the task offloading subproblem constructed in step 6.1, and we get ; represents the updated task offloading decision matrix at the current iteration r+1 steps; fixed , use the interior point method to solve the bandwidth allocation and computing resource allocation sub-problems constructed in step 6.2, and get ; Represents the updated computing resource allocation decision matrix under the current iteration r+1 steps; represents the updated bandwidth allocation decision matrix at the current iteration r+1 steps; fixed , use the heuristic algorithm with polynomial time complexity designed in step 6.4 to solve the service acquisition subproblem constructed in step 6.3, and get ; Represents the updated service acquisition decision matrix at the current iteration r+1 steps; Step 6.5.3, Determination: Determine the value of the objective function Is it equal to , if they are equal, terminate the loop and output the variable matrix ; If not equal, proceed to step 6.5.4; Step 6.5.4, variable update: Updated to ,make , and repeat step 6.5.2; The final approximate optimal decision is a set of matrices obtained when the iterative algorithm converges ,in represents the optimal task offloading decision matrix, represents the optimal computing resource allocation decision matrix, represents the optimal bandwidth allocation decision matrix, Represents the optimal service acquisition decision matrix.

8. The method for low-latency vehicle active service caching and task offloading for Internet of Vehicles according to claim 7, characterized in that: Based on the obtained approximate optimal decision, the vehicle proactively caches services and offloads tasks, including: Task offloading: Based on , each mission vehicle Know your mission Where to uninstall: If , the task is used locally computing resources for processing; if , then the mission vehicle uses Bandwidth offloads tasks to RSU, RSU uses The computing resources for task processing; if a service vehicle , , then the mission vehicle uses Bandwidth offloads tasks to service vehicles , service vehicles use computing resources for task processing; Active service acquisition: When the task Unloaded onto a service vehicle and service vehicles No caching required for serving models When the service vehicle will be based on Active service acquisition process: If , then the service vehicle Actively obtain service models from RSU ;like , then the service vehicle Active slave mission vehicle Get the service model .

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the vehicle active service caching and task offloading method for low latency in the Internet of Vehicles as described in any one of claims 1-8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the vehicle active service caching and task offloading method for low latency in the Internet of Vehicles as described in any one of claims 1-8 are implemented.

Images