ARIS-assisted vehicle-to-everything (V2X) edge computing method and system based on DDPG in the Internet of Vehicles

Abstract

This invention belongs to the field of wireless communication technology and discloses a method and system for ARIS-assisted vehicle-to-everything (V2X) edge computing based on Deep Deterministic Policy Gradient (DDPG). This invention introduces the concept of an unmanned aerial vehicle (UAV) equipped with a reconfigurable smart surface (RIS) as an aerial RIS (ARIS), integrating RIS and UAV into a novel and efficient VEC network auxiliary device. It leverages the flexibility of UAVs and the low cost of RIS to overcome communication performance and cost limitations, assisting vehicles in offloading computation. This invention proposes an algorithm for ARIS-assisted V2X edge computing based on the Deep Deterministic Policy Gradient (DDPG) algorithm. By jointly optimizing the UAV trajectory and the phase shift of the RIS, passive beamforming is achieved, thereby maximizing the completion rate of onboard computing tasks. Numerous numerical results demonstrate that the DDPG-based ARIS-assisted VEC network scheme outperforms other algorithms, improving task completion rate by up to 26%.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and in particular relates to an ARIS-assisted vehicle-to-everything (V2X) edge computing method and system based on DDPG in the Internet of Vehicles (IoV). Background Technology

[0002] With the rapid development and integration of mobile internet and intelligent transportation systems, the role of the Internet of Vehicles (IoV) in the transportation sector is becoming increasingly prominent. The number of connected and autonomous vehicles under the IoV architecture is also gradually increasing with the rapid development of IoV. Meanwhile, driven by big data and artificial intelligence technologies, computationally intensive and latency-sensitive in-vehicle applications (such as real-time traffic analysis, environmental recognition, autonomous driving, video streaming services, and map navigation) have surged at an unprecedented rate in recent years. The continuous emergence of diverse services and new applications has led to a gradual increase in the amount of vehicle mobile data; however, many vehicles currently lack sufficient computing resources, making the problems of computational overload and processing timeouts increasingly serious.

[0003] In recent years, unmanned aerial vehicles (UAVs) have been widely used in various fields due to their high mobility, flexibility, line-of-sight capability, and low cost, such as weather monitoring, forest fire detection, traffic control, cargo transportation, and aerial communication platforms. Among these, UAV-assisted vehicular edge computing (VEC) has attracted significant attention and become a research focus in academia. Compared to traditional fixed-location VECs, UAV-assisted VECs (integrating edge servers into UAVs) offer high flexibility, line-of-sight transmission, and ease of deployment, making them highly useful in handling emergencies, temporary events, and on-demand services. Some studies have proposed using UAVs as computing and relay nodes to reduce the average terminal latency in UAV-assisted VEC systems, establishing a UAV-VEC problem with the goal of minimizing the average latency of all terminals. However, the limited battery capacity of UAVs and the increased energy consumption related to computation and propulsion when installing edge servers on UAVs reduce the UAV's endurance. Furthermore, deploying an additional edge server on the UAV increases the production cost of the UAV.

[0004] To maintain the required transmission in a more reliable manner, an emerging technology called Reconfigurable Intelligent Surface (RIS) is considered a key technology for improving communication performance and enabling intelligent control of the wireless environment in future wireless systems due to its low cost, low power consumption, and tunability. RIS is a surface structure composed of a large number of low-cost passive reflective elements. By adjusting the amplitude and phase shift of the reflective elements, fine-grained reflected beamforming can be achieved. Furthermore, RIS only reflects existing radio frequency signals in the environment without generating new ones, resulting in low power consumption and no new interference. Some studies have deployed RIS on buildings between vehicles and roadside units (RSUs) to allow signals to bypass obstacles between the vehicles and RSUs, avoiding penetration loss in the path and maintaining ideal transmission rates. However, fixed RIS on buildings are limited by the building's layout and structure, making it difficult to modify or reconfigure them after installation. Fixed RIS only have a 180° half-space reflection, limiting signal coverage and causing attenuation or blockage in certain directions, resulting in a significant lack of freedom and flexibility.

[0005] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0006] In UAV-assisted VEC networks, installing edge servers on UAVs increases their weight, leading to increased propulsion energy consumption. Performing computational tasks on UAVs also increases computation-related energy consumption, further shortening the already energy-constrained service time of UAVs. Furthermore, deploying an additional edge server on a UAV increases its production cost. In RIS-assisted VEC networks, traditional fixed RIS systems are typically difficult to modify or reconfigure after installation, and their limited 180° half-space reflection restricts signal coverage and causes attenuation or blockage in certain directions, resulting in a significant lack of freedom and flexibility. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides an ARIS-assisted vehicle-to-everything (V2X) edge computing method and system based on DDPG in the Internet of Vehicles (IoV).

[0008] This invention is implemented as follows: A method for ARIS-assisted vehicle-to-everything (V2X) edge computing based on DDPG in the Internet of Vehicles includes:

[0009] Step 1: Model the network scenario, the ARIS-assisted channel model, and the task offloading model;

[0010] Step 2: Construct optimization objectives, clarifying the goal and constraints of maximizing the overall completion rate of computational tasks for vehicle users;

[0011] Step 3: Based on the scenario and constraints, the problem is transformed into a Markov decision problem, and the state space and action space are modeled. At the same time, a corresponding reward function is designed for the optimization objective.

[0012] Step 4: Based on the DDPG algorithm, design an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm, train the model, and compare the trained model with other benchmark algorithms to verify its performance.

[0013] Furthermore, step 1 specifically includes:

[0014] (1) The number and location of RSUs placed on both sides of the traffic intersection and ground vehicle users are known. ARIS, as a mobile relay auxiliary vehicle, will offload the computing task to the distant RSU. The starting position of ARIS is predetermined and it maintains a constant altitude and fixed speed during flight.

[0015] (2) Define the ARIS-assisted channel model; the channel gain between vehicle v and roadside unit s follows Rayleigh fading and can be expressed as:

[0016]

[0017] Where l is the path loss, Δ is the path loss exponent; λ v，s This represents the distance between vehicle v and roadside unit s. and To accommodate the random scattering components following a Gaussian random distribution, the channel gains from vehicle v to RIS and from RIS to roadside unit s are modeled using a Rice distribution and given by the following equations:

[0018]

[0019] Where, λ v，u and λ u，s Let v represent the distance from vehicle v to RIS and RIS to roadside unit s, respectively, and ξ represent the Rice factor; and The line of sight (LoS) is the direct path. and For non-line of sight (NLoS) paths, the total amount of signal received at roadside unit s can be expressed as:

[0020]

[0021] Among them, P v For the transmission power of communication between all vehicles, n sAdded Gaussian white noise at roadside unit s; the channel gain from vehicle v to roadside unit s is |h v，s +O k h v，u Θh u，s | 2 O k ={0,1} indicates whether the communication process is assisted by RIS, h v，u Θh u，s For cascaded channel communication, h v，s For direct channel communication, the uplink interference signal plus noise ratio from vehicle v to roadside unit s and from roadside unit s to vehicle v in time slot t can be expressed as follows:

[0022]

[0023] Where the numerator represents the signal quantity transmitted from vehicle v to roadside unit s for task k, and the denominator represents the total interference from other vehicle tasks during the same time period, σ 2 It is additive white Gaussian noise in roadside unit s; the transmission rates between vehicle v and roadside unit s and between roadside unit s and vehicle v can be expressed as:

[0024] R v,s =B1log2(1+γ) v，s )

[0025] R s，v =B2log2(1+γ) s，v )

[0026] B1 and B2 are the bandwidths allocated in two different scenarios;

[0027] (3) Define the task unloading model; task latency can be expressed as:

[0028]

[0029] In the VEC system, only the RSU provides computing services, while ARIS provides data forwarding. Therefore, three types of latency will occur during the entire service process: the transmission latency T from task k to the roadside unit s. v，s The computation delay T caused by task k being computed at RSU k，s The transmission delay T between the roadside unit s and the vehicle v is caused by the calculation results being sent back to the vehicle. s，v ;

[0030] In the above formula: c s，k This represents the computing resources allocated from roadside unit s to task k. The computing resources of the roadside units are equally distributed to each task. D k，r The result of the calculation for task k; the delay of task k in time slot t can be expressed as:

[0031]

[0032] Furthermore, step 2 specifically includes:

[0033] (1) Define the optimization objective; by jointly optimizing the UAV trajectory and the phase shift of the RIS, coordinate the task offloading of the vehicle to maximize the overall task completion rate of the vehicle users. The joint optimization problem is as follows:

[0034]

[0035] (2) Determine the constraints;

[0036] The constraints are

[0037]

[0038] Constraint C1 ensures that task k is either directly unloaded or unloaded through ARIS. Constraint C2 requires that the movement distance of ARIS within a time slot cannot exceed δ, where δ is the maximum distance that the UAV can fly in each time slot. Constraint C3 ensures that the total computing resources allocated to the task cannot exceed the computing power of the edge node. Constraints C4 and C5 limit the adjustment range of phase shift and amplitude. RIS is a passive reflection, and the amplitude is fixed at 1.

[0039] Furthermore, step 3 specifically includes:

[0040] (1) Construct the state space; In time slot t, the agent's observation space can be represented as:

[0041] s t ={u(t), v(t), r(t)}

[0042] Where u(t), v(t), and r(t) represent the coordinates of the ARIS, the coordinates of all RSUs, and the coordinates of all vehicles, respectively; in this invention, the entire IoV network system acts as an Agent, which interacts with the environment and makes optimization decisions as the solution to the optimization problem.

[0043] (2) Constructing the action space; In time slot t, the agent's action space can be represented as:

[0044] a t ={α t ,θ(t)}

[0045] Where, α t Let θ(t) be the direction of movement of ARIS, and let θ(t) be a vector representing the phase shift of each unit in the RIS.

[0046] (3) Design the reward function; the immediate reward that the Agent can obtain in time slot t can be expressed as:

[0047]

[0048] The reward function should encourage ARIS to jointly optimize the UAV's trajectory and the RIS's phase shift under constraints (C1) to (C5) in order to maximize the mission completion rate.

[0049] Furthermore, step 4 specifically includes:

[0050] An ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm is constructed to solve the problem.

[0051] (1.1) Initialize the parameters ω, θ, ω′, θ′ of the critic network, actor network, critic target network, and actor target network;

[0052] (1.2) Initialize the experience replay buffer B, batch size M, and target network update frequency C;

[0053] (1.3) for each scene do;

[0054] (1.3.1) Reset the positions of ARIS and all vehicles;

[0055] (1.3.2) for each time slot do;

[0056] (1.3.2.1) Obtain the global environment state space s t ;

[0057] (1.3.2.2) Select the specific action for ARIS in this time slot. And update the position information and the phase shift of the RIS element according to the action;

[0058] (1.3.2.3) Obtain immediate rewards r t And transition to the next environmental state. t+1 ;

[0059] (1.3.2.4) Store the sequence information in the experience replay buffer B, and randomly sample M mini-batches of samples from it;

[0060] (1.3.2.5) Update the weights ω of the Critic network using the mean squared error loss function, and update the weights θ of the Actor network using policy gradient descent;

[0061] (1.3.2.6) Update the weights ω′ and θ′ of the Critic target network and the Actor target network in a soft update manner every C steps;

[0062] (1.3.3) End the for loop;

[0063] (1.4) End the for loop.

[0064] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to execute the ARIS-assisted vehicle-to-everything edge computing method based on DDPG in the vehicle network.

[0065] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the V2X network.

[0066] Another objective of this invention is to provide an information data processing terminal for implementing the ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the V2X network.

[0067] Another objective of this invention is to provide an ARIS-assisted vehicle-to-everything (V2X) edge computing system based on DDPG for use in the Internet of Vehicles (IoV), comprising:

[0068] The modeling module is used to model network scenarios, ARIS-assisted channel models, and task offloading models.

[0069] The target construction module is used to construct optimization targets, which clearly define the goal and constraints of maximizing the overall completion rate of computational tasks for vehicle users;

[0070] The function design module is used to transform the problem into a Markov decision problem based on the scenario and constraints, perform state space and action space modeling, and design reward functions for the optimization objective.

[0071] The model training module is used to build an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm, train the model, and compare the trained model with other benchmark algorithms for performance verification.

[0072] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0073] First, because this invention binds a smart metasurface to a drone as a data forwarding device, it fully combines the high mobility of the drone with the low cost, low power consumption, and low weight advantages of the smart metasurface. Compared to the traditional method of using fixed smart metasurfaces to assist vehicle-to-everything (V2X) edge computing, ARIS can move in real time with changes in traffic flow, offering greater flexibility and providing 360° panoramic reflection, making it easier to avoid signal blockage problems caused by buildings and trees.

[0074] Since this invention optimizes the trajectory of a UAV and the phase shift of a smart metasurface in a dynamic scene, the complexity of the scene is greatly increased because the vehicle moves with different accelerations and the UAV carrying the smart metasurface is also moving dynamically. Furthermore, the research problem of this invention is nonlinear and includes multiple continuous and discontinuous constraint variables, further exacerbating the complexity of the optimization problem and making it difficult for traditional algorithms to solve. This invention transforms the problem into a Markov decision process and uses the DDPG algorithm from deep reinforcement learning to solve it. Compared with greedy heuristic algorithms and other reinforcement learning algorithms, the results show that the algorithm proposed in this invention has the optimal solution.

[0075] In the Internet of Vehicles (IoV), computationally intensive and latency-sensitive in-vehicle applications (such as real-time traffic analysis, environmental recognition, autonomous driving, video streaming services, and map navigation) have enormous demands for computing resources. To enable computational tasks in these scenarios to be completed within tolerable latency, this invention discloses a trajectory and phase shift optimization method for aerial intelligent metasurface-assisted IoV edge computing. This invention can obtain an optimal solution that maximizes the completion rate of computational tasks, reducing the complexity of modeling using traditional algorithms and providing a more efficient solution algorithm for task computation in IoV scenarios. This invention is applicable to various application scenarios such as intelligent transportation systems, autonomous driving systems, and vehicle communication systems.

[0076] This invention addresses the trajectory and phase shift optimization problems of aerial intelligent metasurface-assisted vehicle-to-everything (V2X) edge computing by employing deep reinforcement learning. Compared to traditional optimization algorithms, deep reinforcement learning can better handle complex problems and reduce computational resource and time consumption. This invention addresses the challenges of limited computational resources and constraints such as building occlusion, signal interference, and flight distance in ground vehicle computation task unloading scenarios. To maximize the computation task completion rate, the trajectory and phase shift optimization problem of aerial intelligent metasurfaces is modeled as a hybrid optimization problem. Furthermore, considering the complexity of dynamic scenarios, the proposed optimization problem is modeled as a Markov decision process, and the state space, action space, and reward function are designed. Finally, a UAV trajectory optimization and RIS phase shift optimization strategy based on the DDPG algorithm is designed and implemented.

[0077] In designing the reward function, this invention fully considers the importance of task timeliness and accuracy for later applications. The reward function design provides a larger reward for computation tasks that complete unloading, computation, and data transmission within a tolerable latency. This incentivizes the drone to fly to areas with heavy traffic as quickly as possible, allowing the RIS to adjust the optimal phase shift matrix. This enables vehicle computation tasks to be offloaded to lighter roadside unit nodes, achieving load balancing and maximizing task completion rates.

[0078] Secondly, this invention optimizes the trajectory of the UAV and the phase shift of the RIS (Range Reflection Array) using deep reinforcement learning, which can effectively improve the channel quality of ground vehicles and roadside units, thereby shortening the data transmission time to the edge server and reducing the latency of base station data reception. Simultaneously, optimizing the RIS phase shift enables passive beamforming of reflected signals, allowing computational tasks to be more effectively offloaded to lighter roadside unit nodes, thus avoiding overload of a single edge server in areas with heavy traffic.

[0079] The technical solution protected by this invention addresses the computational offloading optimization problem in the scenario of intelligent metasurface-assisted vehicle-to-everything (V2X) edge computing, and provides a trajectory and phase shift optimization scheme based on deep reinforcement learning. This invention possesses the following technical effects and advantages:

[0080] High-quality channel environment: This invention employs an aerial intelligent metasurface to optimize the channel environment. Through the flexible movement of the UAV and the real-time adjustment of the RIS phase shift matrix, signal transmission is minimized from being affected by buildings and trees. Furthermore, the cascaded channel enhances the signal quality of the direct channel, thereby increasing the data transmission rate.

[0081] Efficient utilization of computing resources: This invention provides an effective solution to the overload phenomenon of edge servers caused by uneven traffic flow distribution at traffic intersections. Through the DDPG-based UAV trajectory and RIS phase shift optimization algorithm proposed in this invention, the aerial intelligent metasurface can adjust its position and phase shift in real time according to the traffic flow distribution, allowing the computing tasks to be evenly unloaded onto multiple roadside unit nodes, thereby maximizing the utilization of the computing resources of each edge server.

[0082] Low operating cost: This invention fully combines the flexibility of drones with the low cost of RIS (Radio Reflectors). Compared to solutions involving drones carrying edge servers and fixed RIS on buildings, the intelligent aerial metasurface has lower production costs, lighter weight, and better energy efficiency. Since the RIS is not fixed, it can adjust its position in real time according to changes in the scene, providing a wider panoramic reflection, and its subsequent maintenance costs can also be effectively controlled.

[0083] Low computational complexity: Compared to traditional optimization algorithms, this invention employs deep reinforcement learning, significantly reducing computational complexity and improving efficiency. Furthermore, the use of the DDPG algorithm for trajectory and phase shift optimization enables rapid identification of the optimal solution, enhancing both unloading and computational efficiency.

[0084] Third, as supplementary evidence of the inventive step of the claims of this invention, it is also reflected in the following important aspects:

[0085] (1) The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0086] The technical solution of this invention addresses the computational offloading optimization problem in the scenario of intelligent metasurface-assisted vehicle-to-everything (V2X) edge computing, and provides a trajectory and phase shift optimization scheme based on deep reinforcement learning. The expected benefits and commercial value of transforming this technical solution into a specific product are mainly reflected in the following aspects:

[0087] Improving the Computation Task Completion Rate: This invention provides an effective solution to the overload phenomenon of edge servers caused by uneven traffic flow distribution at traffic intersections. Through the proposed DDPG-based UAV trajectory and RIS phase shift optimization algorithm, the aerial intelligent metasurface can adjust its position and phase shift in real time according to the traffic flow distribution, allowing the computation task to be evenly unloaded onto multiple roadside unit nodes. This maximizes the utilization of the computing resources of each edge server, thereby improving the computation task completion rate.

[0088] Reduced energy consumption and costs: This invention fully combines the flexibility of drones with the low-cost characteristics of RIS (Radio Reflectors). Compared to solutions involving drones carrying edge servers and fixed RIS on buildings, the intelligent aerial metasurface has lower production costs, lighter weight, and better energy efficiency. Since the RIS is not fixed, it can adjust its position in real time according to changes in the scene, providing a wider panoramic reflection, and its subsequent maintenance costs can also be effectively controlled.

[0089] Improving Task Offloading Quality: This invention employs an aerial intelligent metasurface to optimize the channel environment. Through the flexible movement of the UAV and real-time adjustment of the RIS phase shift matrix, signal transmission is minimized from obstruction by buildings and trees. Furthermore, the cascaded channel enhances the signal quality of the direct channel, thereby improving data transmission rate and task offloading quality.

[0090] In summary, the technical solution of this invention has significant expected benefits and commercial value after transformation. It helps to improve the completion rate of computing tasks, reduce energy consumption and costs, and improve the quality of task unloading, which can bring considerable economic benefits to related industries and enterprises.

[0091] (2) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0092] The technical solution protected by this invention addresses the trajectory and phase shift optimization problem in the scenario of intelligent metasurface-assisted vehicle-to-everything (V2X) edge computing, providing a deep reinforcement learning-based solution that effectively fills a technological gap in the industry both domestically and internationally. Specifically, it is reflected in the following aspects:

[0093] Solving the resource allocation challenge in vehicle-to-everything (V2X) edge computing systems: This technical solution systematically studies and proposes solutions to the problem of uneven load on edge computing servers on both sides of the road in scenarios with dense and unevenly distributed traffic flow, such as road intersections, providing an important foundation and technical support for the development of this field.

[0094] Improving the quality of computational task completion: This invention maximizes the computational task completion rate by optimizing the UAV trajectory and RIS phase shift, enabling the UAV to bypass buildings or trees, reducing signal attenuation caused by obstructions. Simultaneously, the RIS adjusts the phase shift to achieve passive beamforming, enhancing channel gain and minimizing task data transmission latency. Furthermore, this technical solution improves the quality and accuracy of task unloading by dynamically sensing the load status between edge servers and adjusting vehicle unloading decisions.

[0095] Reduced production and usage costs: This invention addresses the high production costs of drones equipped with edge servers and the high maintenance costs of fixed RIS systems, proposing a solution that integrates the RIS system onto the drone. This facilitates flexible deployment of auxiliary equipment, reduces usage costs, and extends service time.

[0096] Fourth, in the Internet of Vehicles (IoV), the efficient offloading and processing of computing tasks for vehicle users is a significant challenge. Traditional methods typically face the following problems: high task offloading latency, uneven allocation of computing resources, low network bandwidth utilization, and severe interference and noise. These problems lead to low completion rates of vehicle user computing tasks, affecting overall network performance and user experience. Furthermore, existing technologies often lack flexibility and adaptability when dealing with dynamically changing network environments, making it difficult to meet the needs of complex application scenarios.

[0097] This invention solves the aforementioned technical problems by designing an Airborne Intelligent Metasurface (ARIS)-assisted vehicle-to-everything (V2X) edge computing method based on the Deep Deterministic Policy Gradient (DDPG) algorithm. First, the network scenario, the ARIS-assisted channel model, and the task offloading model are accurately modeled. By optimizing the objective and constraints for improving the vehicle user's computational task completion rate, the problem is transformed into a Markov decision problem. Through the reasonable setting of the state space, action space, and reward function, efficient offloading and resource allocation of vehicle computational tasks are achieved, significantly improving the computational task completion rate.

[0098] This invention employs the advanced DDPG algorithm to optimize the trajectory and phase shift of ARIS-assisted vehicle unloading calculations. Performance comparisons between the trained model and other benchmark algorithms validate the model's effectiveness and superiority. The DDPG algorithm coordinates vehicle task unloading calculations by jointly optimizing the UAV's trajectory and the RIS's phase shift, achieving efficient resource utilization and task scheduling. The application of this algorithm makes data processing more efficient, decision-making more scientific, and significantly improves task unloading performance in vehicle-to-everything (V2X) networks. Attached Figure Description

[0099] Figure 1 This is a flowchart of the ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the Internet of Vehicles provided in this embodiment of the invention.

[0100] Figure 2 This is a schematic diagram of an application scenario provided by an embodiment of the present invention;

[0101] Figure 3 This is a schematic diagram of a system scenario provided in an embodiment of the present invention;

[0102] Figure 4 This is a flowchart of the solution based on the DDPG algorithm provided in an embodiment of the present invention;

[0103] Figure 5 This is a schematic diagram of a simulation scene provided in an embodiment of the present invention;

[0104] Figure 6 These are simulation result diagrams provided in the embodiments of the present invention;

[0105] Figure 7 This is a block diagram of the ARIS-assisted vehicle-to-everything (V2X) edge computing system based on DDPG in the Internet of Vehicles provided in this embodiment of the invention. Detailed Implementation

[0106] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0107] like Figure 1 As shown in the figure, the ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in a V2X network provided by this invention includes the following steps:

[0108] S101 models the network scenario, the ARIS-assisted channel model, and the task offloading model.

[0109] S102, Construct optimization objectives, clarifying the objective and constraints of maximizing the overall completion rate of computational tasks for vehicle users;

[0110] S103: Based on the scenario and constraints, the problem is transformed into a Markov decision problem, and the state space and action space are modeled. At the same time, a corresponding reward function is designed for the optimization objective.

[0111] S104. Based on the DDPG algorithm, an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm is designed. The model is trained and its performance is verified by comparing it with other benchmark algorithms.

[0112] Step 1 provided in this embodiment of the invention specifically includes:

[0113] (1) The number and location of RSUs placed on both sides of the traffic intersection and ground vehicle users are known. ARIS, as a mobile relay auxiliary vehicle, will offload the computing task to the distant RSU. The starting position of ARIS is predetermined and it maintains a constant altitude and fixed speed during flight.

[0114] (2) Define the ARIS-assisted channel model; the channel gain between vehicle v and roadside unit s follows Rayleigh fading and can be expressed as:

[0115]

[0116] Where l is the path loss, Δ is the path loss exponent; λ v，s This represents the distance between vehicle v and roadside unit s. and To accommodate the random scattering components following a Gaussian random distribution, the channel gains from vehicle v to RIS and from RIS to roadside unit s are modeled using a Rice distribution and given by the following equations:

[0117]

[0118] Where, λ v，u and λ u，s Let v represent the distance from vehicle v to RIS and RIS to roadside unit s, respectively, and ξ represent the Rice factor; and The line of sight (LoS) is the direct path. and For non-line of sight (NLoS) paths, the total amount of signal received at roadside unit s can be expressed as:

[0119]

[0120] Among them, P v For the transmission power of communication between all vehicles, n s Added Gaussian white noise at roadside unit s; the channel gain from vehicle v to roadside unit s is |h v，s +O k h v，u Θh u，s | 2 O k ={0,1} indicates whether the communication process is assisted by RIS, h v，u Θh u，s For cascaded channel communication, h v，s For direct channel communication, the uplink interference signal plus noise ratio from vehicle v to roadside unit s and from roadside unit s to vehicle v in time slot t can be expressed as follows:

[0121]

[0122] Where the numerator represents the signal quantity transmitted from vehicle v to roadside unit s for task k, and the denominator represents the total interference from other vehicle tasks during the same time period, σ 2 It is additive white Gaussian noise in roadside unit s; the transmission rates between vehicle v and roadside unit s and between roadside unit s and vehicle v can be expressed as:

[0123] R v,s =B1log2(1+γ) v，s )

[0124] R s，v =B2log2(1+γ) s，v )

[0125] B1 and B2 are the bandwidths allocated in two different scenarios;

[0126] (3) Define the task unloading model; task latency can be expressed as:

[0127]

[0128] In the VEC system, only the RSU provides computing services, while ARIS provides data forwarding. Therefore, three types of latency will occur during the entire service process: the transmission latency T from task k to the roadside unit s.v，s The computation delay T caused by task k being computed at RSU k，s The transmission delay T between the roadside unit s and the vehicle v is caused by the calculation results being sent back to the vehicle. s，v ;

[0129] In the above formula: c s，k This represents the computing resources allocated from roadside unit s to task k. The computing resources of the roadside units are equally distributed to each task. D k，r The result of the calculation for task k; the delay of task k in time slot t can be expressed as:

[0130]

[0131] S102 provided in this embodiment of the invention specifically includes:

[0132] (1) Define the optimization objective; by jointly optimizing the UAV trajectory and the phase shift of the RIS, coordinate the task offloading of the vehicle to maximize the overall task completion rate of the vehicle users. The joint optimization problem is as follows:

[0133]

[0134] (2) Determine the constraints;

[0135] The constraints are

[0136]

[0137] Constraint C1 ensures that task k is either directly unloaded or unloaded through ARIS. Constraint C2 requires that the movement distance of ARIS within a time slot cannot exceed δ, where δ is the maximum distance that the UAV can fly in each time slot. Constraint C3 ensures that the total computing resources allocated to the task cannot exceed the computing power of the edge node. Constraints C4 and C5 limit the adjustment range of phase shift and amplitude. RIS is a passive reflection, and the amplitude is fixed at 1.

[0138] S103 provided in this embodiment of the invention specifically includes:

[0139] (1) Construct the state space; In time slot t, the agent's observation space can be represented as:

[0140] s t ={u(t), v(t), r(t)}

[0141] Where u(t), v(t), and r(t) represent the coordinates of the ARIS, the coordinates of all RSUs, and the coordinates of all vehicles, respectively; in this invention, the entire IoV network system acts as an Agent, which interacts with the environment and makes optimization decisions as the solution to the optimization problem.

[0142] (2) Constructing the action space; In time slot t, the agent's action space can be represented as:

[0143] a t ={α t ,θ(t)}

[0144] Where, α t Let θ(t) be the direction of movement of ARIS, and let θ(t) be a vector representing the phase shift of each unit in the RIS.

[0145] (3) Design the reward function; the immediate reward that the Agent can obtain in time slot t can be expressed as:

[0146]

[0147] The reward function should encourage ARIS to jointly optimize the UAV's trajectory and the RIS's phase shift under constraints (C1) to (C5) in order to maximize the mission completion rate.

[0148] S104 provided in this embodiment of the invention specifically includes:

[0149] An ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm is constructed to solve the problem.

[0150] (1.1) Initialize the parameters ω, θ, ω′, θ′ of the critic network, actor network, critic target network, and actor target network;

[0151] (1.2) Initialize the experience replay buffer B, batch size M, and target network update frequency C;

[0152] (1.3) for each scene do;

[0153] (1.3.1) Reset the positions of ARIS and all vehicles;

[0154] (1.3.2) for each time slot do;

[0155] (1.3.2.1) Obtain the global environment state space s t ;

[0156] (1.3.2.2) Select the specific action for ARIS in this time slot. And update the position information and the phase shift of the RIS element according to the action;

[0157] (1.3.2.3) Obtain immediate rewards r t And transition to the next environmental state. t+1 ;

[0158] (1.3.2.4) Store the sequence information in the experience replay buffer B, and randomly sample M mini-batches of samples from it;

[0159] (1.3.2.5) Update the weights ω of the Critic network using the mean squared error loss function, and update the weights θ of the Actor network using policy gradient descent;

[0160] (1.3.2.6) Update the weights ω′ and θ′ of the Critic target network and the Actor target network in a soft update manner every C steps;

[0161] (1.3.3) End the for loop;

[0162] (1.4) End the for loop.

[0163] Figure 2 The following provides a more detailed description of specific implementation scenarios of the present invention. The present invention considers establishing an ARIS-assisted VEC network at a traffic intersection, where vehicle computational tasks can be directly offloaded to the nearest RSU or indirectly offloaded to a distant RSU via the ARIS. We assume that S = {S1, S2} RSUs are placed on both sides of the traffic intersection, providing computational offloading services for V = {1, 2, ..., v, ..., V} vehicles. Horizontally placed RISs are bound to UAVs flying at a fixed altitude, providing data forwarding services to ground vehicles. There are a total of N = N RISs. X ×N Y Each unit is connected to a central controller, which dynamically adjusts the phase shift of each unit to optimize signal reflection. Vehicles within the RSU coverage area can choose to offload computational tasks locally or via ARIS to another, more distant RSU, thereby promoting server load balancing and improving task completion rates. The real-world dataset inD-dataset-v1.0 is used in this invention, taking into account vehicle mobility and traffic density. Furthermore, since all vehicles share the same spectrum resources, this invention considers mutual interference between vehicles.

[0164] This invention divides the entire system into T = {1, 2, ..., t, ..., T} time slots, and the system parameters remain constant within each time slot, but differ between different time slots. During time slot t, the coordinates of vehicle v are given by v(t) = {x...} v (t), y v (t), 0} indicates that the random computation task generated by vehicle v is... It means that D k C k γ kThese represent the data size of the computation task, the number of CPU cycles required, and the tolerable latency, respectively. To avoid excessive interference from reflected signals, ARIS can only be used when vehicle v chooses to unload to a distant RSU, with its coordinates being u(t)={x u (t), y u Let H be the vehicle's trajectory and the phase shift of the RIS (t), where H represents a fixed height. Since vehicles are unevenly distributed in urban intersection scenarios, we jointly optimize the drone's trajectory and the RIS phase shift so that vehicles can offload tasks to the correct RSUs to prevent server overload.

[0165] Reference Figure 3 This paper further describes the specific computational task offloading model of the present invention in detail. In the network scenario of the present invention, when a vehicle enters the service range of the RSU, it actively establishes a handshake connection with the RSU through a query to obtain the load status of the RSU. Then, the vehicle will select a suitable RSU for computational offloading based on the server load. Since the computing resources of the edge servers bound to each RSU are limited, and the distribution of vehicles at traffic intersections is usually uneven, the decision to offload to the nearest RSU can lead to server overload and inability to process tasks within the tolerable latency. Therefore, the present invention introduces ARIS for signal reflection to offload computational tasks to RSUs with lower loads. In our invention, only the RSU provides computational services, while ARIS provides data forwarding services. Throughout the entire task lifecycle, only three types of latency will occur: transmission latency during task offloading to the RSU, computational latency during edge server processing, and return latency between the RSU and the vehicle.

[0166] Reference Figure 4 The solution based on the DDPG algorithm of this invention is described in detail below, and the specific algorithm flow is shown in Table 1. In DDPG, the agent's goal is to maximize the expected reward, which can be achieved using... Let γ be the discount factor. Initially, the weights of the four neural networks (θ, θ′, ω, ω′) are randomly initialized, and the experience replay buffer B on the Agent is cleared. Then, the Agent obtains actions from the actor target network based on the state S(t). in Additional noise is introduced to prevent overfitting of the Q-value. ARIS executes action A(t) to update its coordinates, obtaining a new state S(t+1), and receives an immediate reward R(t) in time slot t. To stabilize the training process and improve sample efficiency, the Agent stores the quadruple (S(t), S(t+1), A(t), R(t)) in the experience replay buffer B. Next, a mini-batch of samples {S, S′, A, R} of size M is randomly drawn from the experience replay buffer B to compute the Q-value of the current target network. Based on the obtained Q-value, all parameters of the current Critic network are updated via a neural network using the mean squared error loss function and gradient backpropagation. Then, the Actor network adjusts its parameters by maximizing the output (Q-value) of the Critic network to improve the effectiveness of the policy. Finally, to ensure the stability and convergence of the algorithm, a soft update strategy is adopted, updating the parameters of both the Critic and Actor target networks every C steps.

[0167]

[0168] Table 1

[0169] like Figure 7 As shown in the figure, an ARIS-assisted vehicle-to-everything (V2X) edge computing system based on DDPG in a V2X network provided by an embodiment of the present invention includes:

[0170] The modeling module is used to model network scenarios, ARIS-assisted channel models, and task offloading models.

[0171] The target construction module is used to construct optimization targets, which clearly define the goal and constraints of maximizing the overall completion rate of computational tasks for vehicle users;

[0172] The function design module is used to transform the problem into a Markov decision problem based on the scenario and constraints, perform state space and action space modeling, and design reward functions for the optimization objective.

[0173] The model training module is used to build an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm, train the model, and compare the performance of the trained model with the benchmark algorithm under different ARIS phase shift designs for performance verification.

[0174] To demonstrate the inventiveness and technical value of the technical solution of this invention, this section provides specific product or related technology application examples of the technical solution claimed.

[0175] Example 1: Real-time Traffic Information System

[0176] By deploying edge computing nodes on roads, data generated by vehicle sensors and traffic cameras can be collected and analyzed, and the processed data can then be transmitted to drivers and traffic control centers. This allows drivers to understand local traffic conditions in real time, choose the best routes, and avoid congestion.

[0177] Example 2: Automated Driving System

[0178] Autonomous vehicles are a key application scenario for vehicle-to-everything (V2X) edge computing technology. Autonomous vehicles require real-time analysis of massive amounts of data needed for driving, and V2X edge computing ensures real-time data processing and low latency, thereby guaranteeing the safe and reliable operation of autonomous vehicles. For example, global telecommunications researchers estimate that autonomous vehicles can generate over 40TB of data per hour, which needs to be processed and analyzed through edge computing nodes.

[0179] Example 3: Augmented Reality Navigation System

[0180] Vehicle-to-everything (V2X) edge computing technology can also be applied to augmented reality navigation systems. By using cameras and edge computing nodes on the vehicle, road signs, traffic signals, and the surrounding environment can be identified in real time and overlaid on the driver's field of vision, providing more intuitive and accurate navigation and driving information.

[0181] Example 4: Fleet Management System

[0182] Vehicle-to-everything (V2X) edge computing can be applied to fleet management systems. Cameras on intelligent aerial surfaces can monitor the location, status, and operational status of vehicles in real time, and can dynamically move with changes in traffic flow. This assists the fleet management center in scheduling and managing vehicles, thereby improving traffic efficiency and avoiding congestion on certain road sections.

[0183] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0184] The embodiments of the present invention have achieved some positive results during the research and development or use process, and have indeed great advantages compared with the prior art. The following content describes them in conjunction with the data, charts and other information of the experimental process.

[0185] Figure 5 This is a schematic diagram of the simulation scene of the present invention;

[0186] The invention will be further described below with reference to simulation experiments.

[0187] 1. Simulation conditions:

[0188] The simulation experiment of this invention is conducted on a Windows platform, with the following main configurations: CPU: Intel(R) Core(TM) i5-10400F, 2.90GHz; Memory: 16G; Operating System: Windows 10; Simulation software: PyCharm.

[0189] 2. Simulation content and result analysis:

[0190] This simulation experiment compares the method of this invention with a greedy heuristic algorithm and a TD3-based deep reinforcement learning algorithm to verify and evaluate the superior performance of the DDPG-based ARIS-assisted vehicle computational unloading. The simulation considers a 60m × 90m intersection with unevenly distributed vehicles. Within the intersection, two RSUs associated with the edge server are located on the roadside, and 15 vehicles dynamically move along the road and request computational unloading services from the RSUs. Additionally, a drone attached to the RIS provides data forwarding for the ground vehicles. In the simulation, we set the wireless communication parameters (as shown in Table 2) and the hyperparameters of the DDPG network (as shown in Table 3). In the algorithm model, the neural network architecture includes four fully connected layers with 800, 600, 512, and 256 neurons respectively, using ReLU as the activation function between adjacent layers. The output layer uses the tanh function to normalize the output.

[0191]

[0192] Table 2

[0193]

[0194] Table 3

[0195] Figure 6 Figure A depicts the flight trajectory of the ARIS-assisted vehicle calculating unloading in a load balancing scenario, taking off from a fixed position {20m, -15m, 50m}. The green line represents the ARIS's flight path, the red triangle represents the RSU, and the lines of different colors represent the movement trajectories of different vehicles. It can be seen that the ARIS moves with the changes and distribution of traffic flow, providing data forwarding services for congested road sections.

[0196] Figure 6 B compares the convergence curves of DDPG with other benchmark algorithms. Specifically, the Random algorithm refers to the algorithm that performs random phase shift optimization on ARIS under the same trajectory as the DDPG-based algorithm, while the K-means algorithm generates ARIS trajectories based on vehicle clustering and optimizes the phase shift using the DDPG-based algorithm. We observe that the proposed DDPG-based optimization algorithm has the fastest convergence speed and obtains the highest reward. Considering the complexity of vehicle and ARIS mobility in our scenario, deterministic policies tend to yield more effective action decisions. This makes the DDPG algorithm outperform the TD3 algorithm in our simulation experiments. Furthermore, the TD3 algorithm has a more complex network structure and is more inclined to explore random policies.

[0197] Figure 6The C-squared model compares the computational task completion rates achieved under different computational task sizes. It was observed that the vehicle task completion rate decreases as the amount of computational task data increases. Despite varying task sizes, the proposed DDPG algorithm consistently outperforms other algorithms in terms of task completion rate. Furthermore, as the computational task size increases, the performance gap between DDPG and other algorithms in terms of task completion rate widens. This can be attributed to the fact that even with a less-than-ideal RIS phase shift, a good data transfer rate can still be achieved when the computational task data volume is small. However, when handling larger computational tasks, the benefits of an optimal phase shift become more pronounced, highlighting the impact of the cascaded channel gain brought by ARIS.

[0198] To demonstrate the superiority of ARIS in improving computational task completion rates, Table 4 illustrates the improvement in computational task completion rates achieved by using ARIS in different algorithms, with the nearest unloading algorithm without ARIS serving as the baseline. We observed that the vehicle's task completion rate decreases as the computational task size increases. When the computational task data volume is 3000 bits, the vehicle's computational task completion rate without ARIS is 65.96%. As shown in Table 4, with the help of ARIS, the task completion rates of the four algorithms improved by 19.91%, 24.37%, 25.36%, and 26.23%, respectively.

[0199]

[0200] Table 4

[0201] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for ARIS-assisted vehicle-to-everything (V2X) edge computing based on DDPG in the Internet of Vehicles (IoV), characterized in that, Includes the following steps: Step 1: Model the network scenario, the ARIS-assisted channel model, and the task offloading model; Step 2: Construct optimization objectives, clarifying the goal and constraints of maximizing the overall completion rate of computational tasks for vehicle users; Step 3: Based on the scenario and constraints, the problem is transformed into a Markov decision problem, and the state space and action space are modeled. At the same time, a corresponding reward function is designed for the optimization objective. Step 4: Based on the DDPG algorithm, construct an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm, train the model, and compare the trained model with other benchmark algorithms to verify its performance. Step 2 specifically includes: (1) Determine the optimization objective; By jointly optimizing the UAV's trajectory and the RIS phase shift, and coordinating the task computation offloading of the vehicle, the overall computation completion rate of the vehicle users is maximized. The joint optimization problem is as follows: ， (2) Determine the constraints; The constraints are ， constraint Guarantee the task Either uninstall directly or uninstall via ARIS, with constraints. The requirement is that the ARIS's movement distance within a time slot cannot exceed [a certain limit]. ,in It is the maximum distance that the drone can fly in each time slot, a constraint. Ensure that the total computing resources allocated to the tasks do not exceed the computing power of the edge nodes, thus constraining... and constraints The adjustment range of phase shift and amplitude is limited. RIS is a passive reflection system, and the amplitude is fixed at 1.

2. The ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the Internet of Vehicles as described in claim 1, characterized in that, Step 1 specifically includes: (1) The number and location of RSUs placed on both sides of the traffic intersection and ground vehicle users are known. ARIS, as a mobile relay auxiliary vehicle, will unload the computing task to the distant RSU. The starting position of ARIS is predetermined and it maintains a constant altitude and fixed speed during flight. (2) Define the ARIS-assisted channel model; vehicle and roadside units The channel gain between them follows Rayleigh fading and can be expressed as: ， in, For path loss, The path loss index; Indicates vehicle and roadside units The distance between them and To conform to the random scattering components of a Gaussian random distribution; by vehicle To RIS and RIS to roadside unit The channel gain is modeled using the Rice distribution and is given by the following equations: ， ， in, and Representing vehicles To RIS and RIS to roadside unit distance, Represents Rice factor; and Line of Sight (LoS) and Non-Line of Sight (NLoS); Roadside Unit The total amount of signal received at this location can be expressed as: ， in, For the transmission power of communication between all vehicles, roadside unit Additional Gaussian white noise; from the vehicle To the roadside unit The channel gain is ,in Indicates whether the communication process is assisted by RIS. For cascaded channel communication, For direct channel communication; in Time slot from vehicle To the roadside unit and roadside units vehicle The uplink interference signal plus noise ratio can be expressed as follows: ， ， Among them, the molecule represents the vehicle. Task Transmitted to roadside unit The semaphore represents the total interference from other vehicle tasks during the same time period. It is a roadside unit Additive white Gaussian noise; vehicle To roadside unit s and roadside unit To the vehicle The transmission rates between them can be expressed as: ， ， in, and The bandwidth is allocated in two cases; (3) the task offloading model is defined; the task latency can be expressed as: ， In the VEC system, only RSU provides compute services, while ARIS provides data forwarding. Therefore, the entire service process will generate three types of latency: task latency, task latency, and data latency. Unload to roadside unit Transmission delay at the location ,Task The computation delay generated at RSU. and roadside units The calculation results are sent back to the vehicle. The resulting transmission delay ; In the above formula: Represents roadside unit Assigned to task The computing resources of the roadside units are equally allocated to each task. For the task Calculation results; task exist The time delay within a time slot can be expressed as: 。 3. The ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the Internet of Vehicles as described in claim 1, characterized in that, Step 3 specifically includes: (1) Construct the state space; in time slot t, the agent's observation space can be represented as: ， in, These are represented as the coordinates of ARIS, the coordinates of all RSUs, and the coordinates of all vehicles, respectively. In this invention, the entire IoV network system acts as an Agent, which interacts with the environment and makes optimization decisions as the solution to the optimization problem. (2) Constructing the action space; In time slot t, the agent's action space can be represented as: ， in, This represents the direction of ARIS's movement. Let the phase shift of each unit in the RIS be represented by a vector; (3) Design the reward function; the immediate reward that the Agent can obtain in time slot t can be expressed as: ， The reward function should encourage ARIS to... Under constraints, the trajectory of the UAV and the phase shift of the RIS are jointly optimized to maximize the mission completion rate.

4. The ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in the Internet of Vehicles as described in claim 1, characterized in that, Step 4 specifically includes: An ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm is constructed to solve the problem. (1.1) Initialize the parameters of the critic network, actor network, critic target network, and actor target network. ; (1.2) Initialize the experience playback buffer Batch size and the frequency of updating the target network ; (1.3) for each scene do; (1.3.1) Reset the positions of ARIS and all vehicles; (1.3.2) for each time slot do; (1.3.2.1) Obtain the global environment state space ; (1.3.2.2) Select the specific action for ARIS in this time slot. And update the position information and phase shift of the RIS element according to the action; (1.3.2.3) Receive instant rewards And transition to the next environmental state ; (1.3.2.4) Store the sequence information in the experience replay buffer. From which random samples are taken. A small batch of samples; (1.3.2.5) Update the weights of the Critic network using the mean squared error loss function. Update the weights of the Actor network using policy gradient descent. ; (1.3.2.6) Every The weights of the Critic and Actor target networks are updated using a soft update method. ; (1.3.3) End the for loop; (1.4) End the for loop.

5. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor causes the processor to perform the ARIS-assisted vehicle-to-everything edge computing method based on DDPG in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the ARIS-assisted vehicle-to-everything (V2X) edge computing method based on DDPG in any one of claims 1 to 4.

7. A vehicle-to-everything (V2X) system based on DDPG-based ARIS-assisted V2X edge computing method according to any one of claims 1 to 4, characterized in that, The ARIS-assisted vehicle-to-everything (V2X) edge computing system based on DDPG in the aforementioned vehicle-to-everything (V2X) network includes: The modeling module is used to model network scenarios, ARIS-assisted channel models, and task offloading models. The target building module is used to construct optimization targets, which clearly define the goal and constraints of maximizing the overall completion rate of computational tasks for vehicle users. The function design module is used to transform the problem into a Markov decision problem based on the scenario and constraints, perform state space and action space modeling, and design reward functions for the optimization objective. The model training module is used to build an ARIS-assisted vehicle unloading trajectory and phase shift optimization algorithm based on the DDPG algorithm, train the model, and compare the trained model with other benchmark algorithms for performance verification.

Images