Link Topology Adaptive Offloading Method for Edge Computing in Vehicle-to-Everything (V2X)

Abstract

This invention discloses a link topology adaptive offloading method for edge computing in the Internet of Vehicles (IoV). A dynamic V2I and V2V joint system is constructed, and the latency minimization problem of sequential subtasks in the joint system is formulated as a path optimization problem. A Markov decision process (DDQN) is defined, including its state space, state at each time step, action space, and reward. The path optimization problem is modeled as a Markov decision process. The Markov decision process utilizes a Generative Network (GCN) to adaptively learn hidden information in the state at different time steps based on different link topologies, and provides this information to a Direct Decision-Making Network (DDQN) to select the action that yields the highest reward, thereby minimizing the total latency. This invention develops a fine-grained partial offloading model for sequential subtasks in a dynamic V2I and V2V joint system with link topologies to optimize the cooperative offloading strategy. Based on GCN and DDQN adaptive offloading methods, it adapts to different link topologies and minimizes the total latency.

Description

Technical Field

[0001] This invention belongs to the fields of reinforcement learning and graph machine learning technology, specifically a link topology adaptive offloading method for edge computing in vehicle-to-everything (V2X) networks. Background Technology

[0002] In recent years, with the development of communication technology, some innovative models have emerged, such as vehicle-to-vehicle (V2V) communication and vehicle-to-infrastructure (V2I) communication, which establish direct and decentralized links between nearby vehicles. Through these communications, vehicles connect with infrastructure elements and exchange information. These models enable vehicles to effectively exchange real-time information, laying the foundation for numerous applications such as collision avoidance, cooperative driving, and traffic management.

[0003] In high-demand scenarios such as traffic congestion or emergencies, vehicles can pool computing power and coordinate their efforts to handle increased loads, with nearby vehicles and infrastructure sharing computing resources. Cluster-based algorithms treat buses as cluster heads, overcoming resource constraints and meeting the Quality of Service (QoS) requirements for emergency information. Collision warning systems at unsignalized intersections based on V2V and V2I can improve driver safety. Some algorithms model vehicle-to-roadside-units (RSUs) and vehicle connectivity, estimating vehicle connectivity and unloading capacity through traffic classification, overload control, and access control. Other algorithms optimize transmission patterns and power levels in V2V and V2I environments to maximize the total capacity of V2I links while ensuring QoS requirements. Clearly, V2V and V2I collaborative computing paradigms effectively utilize resources, improve response time, and enhance overall system performance.

[0004] However, the rapid movement of vehicles leads to instability in communication links and dynamic changes in network topology, severely impacting data transmission performance and hindering the development of the Internet of Vehicles (IoV). Some methods select serving vehicles by considering link duration and distance between nodes, utilizing idle resources on vehicle nodes to offload computational tasks. Other methods select vehicles with nearby available idle computing resources by considering factors such as link reliability, distance, available computing resources, and relative speed to achieve concurrent task processing. Furthermore, V2V communication raises concerns about potential privacy violations and unauthorized exposure of personal data, potentially leading some vehicle owners to choose not to participate in V2V communication, thus preventing some vehicles from establishing connections with other vehicles. Therefore, to achieve reliable communication, the links between vehicles must form a specific topology to ensure communication quality and privacy security. Link reliability is closely related to various factors. The link topology formed under the influence of indicators constructed based on these different factors will also vary.

[0005] In practice, determining which vehicles provide computing resources based solely on statistical information about resource demand is impractical, as choices made based solely on statistics cannot adapt to changing demands over time. Therefore, many studies utilize reinforcement learning to optimize offloading strategies in IoT environments, as reinforcement learning allows vehicles to learn and optimize their communication strategies based on environmental penalties and rewards. By employing reinforcement learning techniques, vehicles can learn how to dynamically optimize important factors such as transmission power, channel allocation, and resource allocation, thereby achieving more efficient and reliable V2V communication. Unfortunately, existing research neglects the highly dynamic nature of the topological connections between vehicles, which is crucial in real-world V2V communication scenarios.

[0006] Explanation of terms:

[0007] V2V: Vehicle-to-Vehicle, communication between vehicles;

[0008] V2A: Vehicle-to-Assist vehicle, communication between a vehicle and its assistance vehicle;

[0009] V2I: Vehicle-to-Infrastructure (V2I) communication between vehicles and infrastructure. Summary of the Invention

[0010] The purpose of this invention is to address the offloading decision problem in joint vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems with different link topologies. A link topology-adaptive offloading method for edge computing in vehicle-to-everything (V2X) networks, namely Link Topology-Adaptive Offloading (LTAO), is proposed to adaptively offload tasks according to different link topologies in joint V2V and V2I systems. It minimizes the total latency of joint V2V and V2I systems composed of multiple mobile vehicles. To find the optimal offloading strategy under different link topologies, the offloading process of the topology-aware part in the joint V2V and V2I system is modeled as a Markov Decision Process (MDP), and the optimization problem is solved using Graph Convolutional Networks (GCN) and Double Deep-Q Network (DDQN).

[0011] In a first aspect, the present invention provides a link topology adaptive offloading method for edge computing in vehicle-to-everything (V2X) networks, comprising the following steps:

[0012] Step 1: Obtain the data source, including the topology link between the vehicle and the intelligent roadside device, the data transmission rate and task computing capability, as well as the vehicle's task data volume, task computing workload, coordinates, speed and task start time;

[0013] Step 2: Build a joint system of vehicle-to-vehicle and vehicle-to-infrastructure, and then formulate the problem of minimizing the latency of sequential subtasks in the joint system as a path optimization problem;

[0014] The vehicle-to-vehicle and vehicle-to-infrastructure joint system includes multiple ordinary vehicles V1, V2, ..., V M A vehicle V0 and a smart roadside device, where M is an integer greater than 1;

[0015] Ordinary vehicles selectively communicate with other ordinary vehicles to form V2V links, auxiliary vehicles communicate with any ordinary vehicle to form V2A links, and intelligent roadside equipment communicates with any vehicle to form V2I links. The auxiliary vehicle is responsible for generating tasks and has the ability to execute its own designated tasks and unload tasks from other networked ordinary vehicles. The ordinary vehicle is responsible for generating tasks and has the ability to execute its own designated tasks and unload tasks from other networked ordinary vehicles. The intelligent roadside equipment has the ability to execute tasks from the auxiliary vehicle itself and unload tasks from ordinary vehicles.

[0016] Step 3: Define the state space NF of the Markov decision process.p The path optimization problem is modeled as a Markov decision process, consisting of the state, action space A, and reward at each time step.

[0017] Markov decision processes utilize graph convolutional neural networks (GCNs) to adaptively learn the state S at time steps based on different link topologies. p The hidden information is provided to the Deep Dual Q Network (DDQN) to select the action that yields the highest reward in order to minimize the total latency.

[0018] In a second aspect, the present invention provides an electronic device, characterized in that it includes a processor and a memory, the memory storing machine-executable instructions executable by the processor, the processor executing the machine-executable instructions to implement the method.

[0019] Thirdly, the present invention provides a machine-readable storage medium, characterized in that the machine-readable storage medium stores machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method.

[0020] The beneficial effects of this invention are as follows:

[0021] This invention develops a fine-grained partial offloading model for sequential subtasks in dynamic V2I and V2V joint systems with topological links to optimize cooperative offloading strategies, thereby achieving efficient utilization of computing resources. Furthermore, this invention proposes a link topology adaptive offloading method based on GCN and DDQN to adapt to different link topologies and minimize total latency. Attached Figure Description

[0022] Figure 1 This is the overall architecture of the V2V and V2I combined system proposed in this invention.

[0023] Figure 2 These are some possible structural patterns of topological links when the number of vehicles M=4.

[0024] Figure 3 This is a schematic diagram of the TLAO-DDQN link topology adaptive offloading method for edge computing in the Internet of Vehicles proposed in this invention.

[0025] Figure 4 It is the average delay of different methods under different numbers of vehicles, where the number of vehicles M = [5, 6, 7, 10, 15].

[0026] Figure 5 This is a comparison of TLAO-DDQN with a greedy algorithm and a greedy algorithm that does not divide into subtasks, where the number of vehicles M = [5, 6, 7, 10, 15]. Detailed Implementation

[0027] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0028] Offloading methods play a crucial role in optimizing computation, minimizing latency, and enhancing the overall performance of vehicular edge computing (VEC) systems by transferring tasks between vehicles and edge servers or other computing resources. However, many works neglect the topological links that arise due to differences in privacy requests, communication preferences, and transmission quality, which significantly impact the search space for optimal offloading decisions. This invention proposes a graph reinforcement learning (GRL) method to adaptively optimize the sum of latency for tasks in joint vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, termed the linktopology-adaptive offloading (LTAO) method. The LTAO method consists of two concatenated modules: a graph convolutional network (GCN) module and a deep reinforcement learning (DRL) module. The former extracts features from the current state, while the latter outputs the offloading decision based on the state. Extensive experimental results demonstrate the effectiveness and superiority of the proposed LTAO method.

[0029] Therefore, this embodiment provides a LinkTopology-Adaptive Offloading (LTAO) method for edge computing in the Internet of Vehicles (IoV), such as... Figure 3 As shown, this includes modeling the joint V2V and V2I system, the input and output of the DDQN model, and the training process. First, the task is divided into K sequential subtasks, and subtask information and topological link information from the environment are input as node features and adjacency matrices. Next, the node features and adjacency matrix are used as state information to learn new hidden representations through a GCN. The output of the GCN is then used as input to the DDQN for the next action selection learning to make the optimal choice and update the environment at the next time step.

[0030] The method includes the following steps:

[0031] Step 1: Obtain the data source, including the topology link between the vehicle and the intelligent roadside device, the data transmission rate R and the task computing power f, as well as the vehicle's task data volume D, task computing workload, coordinates (x, y), speed v and task initial time T. stsrt ;

[0032] The intelligent roadside equipment refers to the roadside unit (RSU);

[0033] Step 2: Build a joint system of vehicle-to-vehicle and vehicle-to-infrastructure, then formulate the problem of minimizing the latency of sequential subtasks in the joint system as a path optimization problem, and model it as a Markov decision process;

[0034] like Figure 1 As shown, the vehicle-to-vehicle and vehicle-to-infrastructure joint system includes multiple ordinary vehicles V1, V2, ..., V M A vehicle V0 and a smart roadside device, where M is an integer greater than 1;

[0035] Ordinary vehicles selectively communicate with other ordinary vehicles to form V2V links; auxiliary vehicles communicate with any ordinary vehicle to form V2A links; and intelligent roadside equipment communicates with any vehicle (referring to both ordinary and auxiliary vehicles) to form V2I links. The auxiliary vehicles are responsible for generating tasks and have the ability to execute their own designated tasks and unload tasks from other networked ordinary vehicles. Ordinary vehicles are responsible for generating tasks and have the ability to execute their own designated tasks and unload tasks from other networked ordinary vehicles. The intelligent roadside equipment has the ability to execute tasks originating from the auxiliary vehicles themselves and unload tasks from ordinary vehicles.

[0036] The specific task unloading relationship is: to unload vehicle V m The task is divided into K subtasks, where m∈{0,1,2,…,M}. Subtasks related to the same vehicle are arranged sequentially, and subsequent subtasks are generated based on previous subtasks. The initial task arrival time assigned to each ordinary vehicle is random. Indicates vehicle V m The arrival time of the task;

[0037] For each vehicle V m Its sub-task set is represented as Task m ={task m,1 ,task m,2 ,…,task m,K Specifically, vehicle V m The first subtask m,1It can be executed within the vehicle itself, or it can be offloaded to intelligent roadside equipment, auxiliary vehicle V0, or other ordinary communication vehicles; the last subtask m,K After successful completion, the results need to be transmitted back to vehicle V. m ;

[0038] The latency minimization problem of sequential subtasks in V2I and V2V joint systems consists of the following three parts: i) execution time, which depends on the amount of computation and the computing power of the device; ii) transmission time, which is calculated based on the data size and transmission rate (if a link exists); and iii) waiting time, which is determined by the offloading strategy of the previous subtasks and the limitations of available computing resources.

[0039] 1) Execution time:

[0040] Assuming Dev0 corresponds to the auxiliary vehicle V0, Dev M+1 Corresponding intelligent roadside devices, Dev1, Dev2, ..., Dev M These correspond to ordinary vehicles V1, V2, ..., V respectively. M Therefore, it is denoted as device Dev. l (where l∈{0,1,…,M+1}; device Dev) l They all have the ability to execute the subtasks assigned to them;

[0041] subtask m,k The execution time is expressed as:

[0042]

[0043] in This indicates that for ordinary vehicles V m The kth subtask m,k Device Dev l The execution time; exe represents the execution process; ω m,k Indicates subtask m,k The computational workload, m∈{0,1,2,…,M},k∈{1,2,…,K}; fl represents the device Dev l The computing power, measured in cycles per second;

[0044] Used to determine the task m,K Is it in device Dev? l The above is executed, and the calculation formula is as follows:

[0045]

[0046] 2) Transmission time:

[0047] Intelligent roadside devices and auxiliary vehicles are capable of communicating with all ordinary vehicles; however, due to reasons such as privacy, trust authorization, and communication intent, connectivity between ordinary vehicles is limited; these limitations manifest as masks. m =[mask m,1 ,mask m,2 ,…,mask m,M ];

[0048]

[0049] Where m'∈{1,2,…,M}, m≠m';

[0050] task m,k Not always from vehicle V m Transmission, unless k=1. Introducing CD. m,k Indicates the task currently responsible for performing the task. m,k The equipment, namely:

[0051] When k=1, vehicle V m Execute subtasks by itself m,k ;

[0052] When k > 1, execute the subtask. m,k The device is denoted as CD. m,k Therefore, subtask m,k Transmission time The calculation formula is:

[0053]

[0054] Where, cur represents CD m,k D m,k Indicates subtask m,k The data size is represented by tr, which indicates the transmission process. Indicates subtask m,k Whether to transmit to the target device Dev l The indicated value; mask cur,l Specify from CD m,k To Dev l Does the link exist?

[0055] Current device CD m,k To device Dev l transmission rate R cur,l The calculation formula is:

[0056]

[0057] Where h is the channel fading coefficient, P is the transmission power, θ is the path loss exponent, and ω0 is the white Gaussian noise power; B cur,l and d cur,l These represent the current device CD. m,k With device Dev l Bandwidth and distance between them;

[0058] 3) Waiting time:

[0059] When a vehicle or smart roadside device is assigned to the current subtask m,k When a task is occupied, other subtasks that have been unloaded onto the occupied device need to wait for the subtask task. m,k Finish;

[0060] When subtask m,k Select on device Dev l Waiting time during execution The calculation is as follows:

[0061]

[0062] in, It is a device Dev l The time occupied, K m It is vehicle V m The number of completed subtasks. It is a task m,k Arrival at device Dev l Arrival time, ΔT m,k,l In the device Dev l Tasks executed on m,k Total delay; K m It is vehicle V m The number of completed subtasks. Indicates vehicle V m The time when subtasks begin to be generated;

[0063] The latency minimization problem of a joint V2I and V2V system is formulated as a path optimization problem, taking into account the possibility of unloading tasks based on task order and subtask dependencies; the problem is formulated as follows:

[0064]

[0065] Where T sum This represents the total latency for executing all subtasks across all devices;

[0066] Finally, the path optimization problem is modeled as a Markov decision process.

[0067] Step 3: Based on Step 2, define the state space NF of the Markov decision process.p and the state at each time step;

[0068] In a V2I and V2V joint system, the time step t p State S p Defined as:

[0069] S p =[NF p A p Equation (11)

[0070]

[0071] Among them, A p It is time t p The adjacency matrix of the topology links; It is V m At time t p The current position in, and It is V m speed;

[0072] Step 4: Based on Steps 2 and 3, define the action space A of the Markov decision process. The action selection at each time step is given by the reinforcement learning method. The reinforcement learning method adopts the deep dual Q network DDQN from Step 6.

[0073] The action space is A = [a0, a1, ..., a M+1 ], where a(t p ) = a0 represents time t p Subtask Executed on the auxiliary vehicle, a(t) p )=[a1,…,a M ] indicates time t p Subtask Executed on a regular vehicle, a(t) p ) = a M+1 Represents time t p Subtask Executed on intelligent roadside equipment;

[0074] Step 5: Based on steps 2, 3, and 4, define the reward for the Markov decision process. Define different rewards and penalties according to different situations. The core objective is to minimize the total delay.

[0075]

[0076] Condition 1 is defined as follows: If a(t) p Not here In the middle, and at time t p Task number k p =K, then the task The result is preferentially transmitted back to the vehicle m p ∈ {0, 1, 2, …, M}, k p ∈ {1, 2, …, K}; represents the current device CD p at time t p and the connection status of other devices;

[0077] Condition 2 is defined as follows: If a(t p ) is not in and k p < K, then the task is preferentially transmitted to the connected device in ;

[0078] where C1, C2, C3 are three different constants used to adjust penalties and rewards according to the preferences of the optimization problem; represents the execution time of the k p -th sub-task a(t p ) of an ordinary vehicle at time t a(t p ) at time t represents the transmission time of the k p -th sub-task a(t p ) of an ordinary vehicle; a(t p ) when it is selected to be executed; represents the waiting time p for an ordinary vehicle at time t p for the k -th sub-task p ;

[0079] Step 6: Based on Steps 2, 3, 4, and 5, use the graph convolutional neural network GCN to adaptively learn the hidden information of the time step state S p in and provide it to the deep double Q-network DDQN to select the action that obtains the highest reward to minimize the total delay of the system; specifically:

[0080] 6-1 Input the state information S p = [NF p , A p into the two-layer graph convolutional neural network GCN to learn the hidden information:

[0081]

[0082] where hp t p The hidden information at time step is represented by W0 and W1, which are weight matrices. It is a symmetric normalized adjacency matrix;

[0083] h p The Q-value is obtained by inputting a two-layer multi-layer perceptron (MLP):

[0084] Q p =MLP(h p =W3ReLU(W2h) p Equation (17)

[0085] Q p Represents time t p The Q value; W2 and W3 are the weight matrices of the MLP;

[0086] The Deep Dual-Q Network (DDQN) described in 6-2 includes a master network and a target network;

[0087] The target network determines the action a at the current moment. p Output Q T ;

[0088] The Q function of the main network is updated as follows:

[0089]

[0090] Where r p Through t p The reward is measured by the time delay in h. p+1 and a p+1 These are the hidden information representation and action for the next time step, Q. M (t p+1 ) as the main network in t p+1 Q-value prediction at time Q T It is the output of the target network; Let time t p The Q function of the main network; This represents the Q function of the updated main network;

[0091] From the ability to obtain the maximum of Output a p+1 That is, at time t p+1 The optimal uninstallation scheme is selected based on the chosen action. The task execution process can then be determined from the action selected based on the optimal uninstallation scheme.

[0092] Figure 2These are some possible connection patterns of the topology when the number of vehicles M=4, which constitute different adjacency matrices and masks.

[0093] Figure 4 This represents the average delay for different methods under different numbers of vehicles, where the number of vehicles M = [5, 6, 7, 10, 15]. For example... Figure 4 As shown, the results of the TLAO-DDQN method of this invention are significantly better than other methods. For example, when the experiment is set to M=5 and K=5, TLAO-DDQN improves performance by 3.71% compared to TLAO-DQN, by 16.9% compared to Greedy, by 16.9% compared to Greedy-noseg, by 58.49% compared to All Edge, by 66.26% compared to All Local, and by 71.76% compared to Random. This is because as M increases, the available equipment in the vehicle also increases, thereby expanding the optimization space for the mean time. In non-DRL-based algorithms, the greedy algorithm achieves the optimal unloading decision, while the DRL-based algorithms TLAO-DDQN and TLAO-DQN consistently outperform non-DRL-based algorithms under different M values. As the number of available equipment increases, the optimization space also expands, and the superiority of the DRL-based method becomes obvious, indicating that they can effectively handle various unloading scenarios in V2I and V2V systems.

[0094] Figure 5 This compares TLAO-DDQN with a greedy algorithm and a greedy algorithm that does not divide tasks, where the number of vehicles M = [5, 6, 7, 10, 15]. Figure 5 As shown, TLAO-DDQN significantly outperforms the two different greedy algorithms. For example, when the experiment is set to K=5, corresponding to different M=[5,6,7,10,15], TLAO-DDQN improves performance by 29.81%, 16.90%, 10.05%, 11.62%, and 10.60% compared to Greedy; and by 36.47%, 38.63%, 32.90%, 18.99%, and 18.96% compared to Greedy-noseg. Since the computational power of RSU remains constant, the computational power of V2V and V2I becomes insufficient as M increases, and the performance improvement of TLAO decreases with increasing M, which means a reduction in the optimizable space.

Claims

1. A link topology adaptive offloading method for edge computing in vehicle-to-everything (V2X) networks, characterized by: The method includes the following steps: Step 1: Obtain the data source, including the topology link between the vehicle and the intelligent roadside device, the data transmission rate and task computing capability, as well as the vehicle's task data volume, task computing workload, coordinates, speed and task start time; Step 2: Build a joint system of vehicle-to-vehicle and vehicle-to-infrastructure, and then formulate the problem of minimizing the latency of sequential subtasks in the joint system as a path optimization problem; The vehicle-to-vehicle and vehicle-to-infrastructure joint system includes multiple ordinary vehicles V1, V2, ..., V M A vehicle V0 and a smart roadside device, where M is an integer greater than 1; Ordinary vehicles selectively communicate with other ordinary vehicles to form V2V links, auxiliary vehicles communicate with any ordinary vehicle to form V2A links, and intelligent roadside equipment communicates with any vehicle to form V2I links. The auxiliary vehicle is responsible for generating tasks and has the ability to execute its own designated tasks and unload tasks from other networked ordinary vehicles. The ordinary vehicle is responsible for generating tasks and has the ability to execute its own designated tasks and unload tasks from other networked ordinary vehicles. The intelligent roadside equipment has the ability to execute tasks from the auxiliary vehicle itself and unload tasks from ordinary vehicles. Step 3: Define the state space NF of the Markov decision process. p The path optimization problem is modeled as a Markov decision process, consisting of the state, action space A, and reward at each time step. Markov decision processes utilize graph convolutional neural networks (GCNs) to adaptively learn the state S at time steps based on different link topologies. p The hidden information in the data is provided to the Deep Dual Q Network (DDQN) to select the action that yields the highest reward, thereby minimizing the total latency; specifically: The status information S p =[NF p A p Input a two-layer graph convolutional neural network (GCN) to learn hidden information: Where h p For t p The hidden information at time step is represented by W0 and W1, which are weight matrices. It is a symmetric normalized adjacency matrix; Among them, A p It is time t p The adjacency matrix of the topology links; It is V m At time t p The current position in, and It is V m speed; h p The Q-value is obtained by inputting a two-layer multilayer perceptron (MLP): Q p =MLP(h p =W3ReLU(W2h) p Equation (17) Q p Represents time t p The Q value; W2 and W3 are the weight matrices of the MLP.

2. The method according to claim 1, characterized in that... The intelligent roadside equipment refers to the roadside unit (RSU).

3. The method according to claim 1, characterized in that... In joint vehicle-to-vehicle and vehicle-to-infrastructure systems, the task offloading relationship specifically involves: offloading tasks from vehicle V... m The task is divided into K subtasks, where m∈{0,1,2,…,M}. Subtasks related to the same vehicle are arranged sequentially, and subsequent subtasks are generated based on previous subtasks. The initial task arrival time assigned to each ordinary vehicle is random. Indicates vehicle V m The arrival time of the task.

4. The method according to claim 3, characterized in that... In joint vehicle-to-vehicle and vehicle-to-infrastructure systems, for each vehicle V m Its sub-task set is represented as Task m ={task m,1 ,task m,2 ,…,task m,K Specifically, vehicle V m The first subtask m,1 It can be executed within the vehicle itself, or it can be offloaded to intelligent roadside equipment, auxiliary vehicle V0, or other ordinary communication vehicles; the last subtask m,K After successful completion, the results need to be transmitted back to vehicle V. m .

5. The method according to claim 3, characterized in that... In joint vehicle-to-vehicle and vehicle-to-infrastructure systems, the latency minimization problem of sequential subtasks consists of three parts: execution time, transmission time, and waiting time. The latency minimization problem of joint systems is formulated as a path optimization problem, taking into account the possibility of unloading tasks based on the dependencies of sequential subtasks.

6. The method according to claim 5, characterized in that... The specific process for obtaining the execution time is as follows: Assuming Dev0 corresponds to the auxiliary vehicle V0, Dev M+1 Corresponding intelligent roadside devices, Dev1, Dev2, ..., Dev M These correspond to ordinary vehicles V1, V2, ..., V respectively. M Therefore, it is denoted as device Dev. l Where l∈{0,1,…,M+1}; device Dev l They all have the ability to execute the subtasks assigned to them; subtask m,k The execution time is expressed as: in This indicates that for ordinary vehicles V m The kth subtask m,k Device Dev l The execution time; exe represents the execution process; ω m,k Indicates subtask m,k The computational workload, m∈{0,1,2,…,M},k∈{1,2,…,K}; f l Indicates device Dev l The computing power, measured in cycles per second; Used to determine the task m,K Is it in device Dev? l The above is executed, and the calculation formula is as follows: The specific process for obtaining the transmission time is as follows: Intelligent roadside devices and auxiliary vehicles are capable of communicating with all ordinary vehicles; however, connectivity between ordinary vehicles is limited due to privacy, trust authorization, and communication intent considerations; these limitations manifest as masks. m =[mask m,1 ,mask m,2 ,…,mask m,M ]; Where m'∈{1,2,…,M}, m≠m'; When k=1, vehicle V m Execute subtasks by itself m,k ; When k > 1, execute the subtask. m,k The device is denoted as CD. m,k Therefore, subtask m,k Transmission time The calculation formula is: Where, cur represents CD m,k D m,k Indicates subtask m,k The data size is represented by tr, which indicates the transmission process. This represents a subtask. m,k Whether to transmit to the target device Dev l The indicated value; mask cur,l Specify from CD m,k To Dev l Does the link exist? Current device CD m,k To device Dev l transmission rate R cur,l The calculation formula is: Where h is the channel fading coefficient, P is the transmission power, θ is the path loss exponent, and ω0 is the white Gaussian noise power; B cur,l and d cur,l These represent the current device CD. m,k With device Dev l Bandwidth and distance between them; The specific process for obtaining the waiting time is as follows: When a vehicle or smart roadside device is assigned to the current subtask m,k When a task is occupied, other subtasks that have been offloaded to the occupied device need to wait for the subtask task. m,k Finish; When subtask m,k Select on device Dev l Waiting time during execution The calculation is as follows: in, It is a device Dev l The time occupied, K m It is vehicle V m The number of completed subtasks. It is a task m,k Arrival at device Dev l Arrival time, ΔT m,k,l In the device Dev l The task executed on m,k Total delay, Indicates vehicle V m The time when subtasks begin to be generated; The problem of minimizing the latency of a joint system can be formulated as a path optimization problem, as follows: Where T sum This represents the total latency for executing all subtasks across all devices.

7. The method according to claim 1, characterized in that... The state space NF of the Markov decision process in step 3 p The specific state at each time step is as follows: In joint vehicle-to-vehicle and vehicle-to-infrastructure systems, the time step t p State S p Defined as: S p = [NF p , A p Formula (11) The action space A = [a0, a1, ..., a] of a Markov decision process M+1 ], where a(t p ) = a0 represents time t p Subtask Executed on the auxiliary vehicle, a(t) p )=[a1,…,a M ] indicates time t p Subtask Executed on a regular vehicle, a(t) p ) = a M+1 Represents time t p Subtask Executed on intelligent roadside equipment; The specific rewards of a Markov decision-making process are: in C1, C2, and C3 are three different constants used to adjust the penalties and rewards according to the preferences of the optimization problem. Represents time t p For ordinary vehicles The kth p Sub-tasks a(t p ) execution time, Represents time t p For ordinary vehicles The kth p Sub-tasks a(t) p Transmission time, Represents time t p For ordinary vehicles The kth p Sub-tasks Choose in a(t) p The waiting time when executing on ) 8. The method according to claim 7, characterized in that... In step 3, the Deep Dual Q Network (DDQN) mentioned in the Markov decision process includes a master network and a target network. The target network determines the action a at the current moment. p Output Q T ; The Q function of the main network is updated as follows: Where r p Through t p The reward is measured by the time delay in h. p+1 and a p+1 These are the hidden information representation and action for the next time step, Q. M (t p+1 ) as the main network in t p+1 Q-value prediction at time Q T It is the output of the target network; Let time t p The Q function of the main network; This represents the Q function of the updated main network; From the ability to obtain the maximum of Output a p+1 That is, at time t p+1 The action to select the optimal uninstallation solution.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the method of any one of claims 1-8.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores machine-executable instructions that, when invoked and executed by a processor, cause the processor to implement the method of any one of claims 1-8.

Images