Resource management method combining vehicle formation control and task unloading and related equipment

Abstract

The invention discloses a resource management method combining vehicle formation control and task unloading and related equipment, and relates to the technical field of Internet of Vehicles. The method comprises the steps that a multi-target and multi-target non-convex constraint optimization problem of a vehicle formation is defined based on driving information, and the multi-target and multi-target non-convex constraint optimization problem is used for enabling task processing time delay and vehicle spacing of the vehicle formation to be minimized under the constraint condition; solving the multi-objective non-convex constraint optimization problem to obtain optimization parameters; and carrying out resource management on the vehicle formation based on the optimization parameters. By implementing the technical scheme provided by the invention, the optimization of the overall performance of the combined vehicle formation can be realized on the premise of ensuring the communication quality.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle networking, and in particular to a resource management method and related equipment for combined vehicle formation control and task offloading. Background Art

[0002] With the rapid development of Internet of Vehicles (IoV) technology, vehicle platoon control technology has become an important research area in intelligent transportation systems. In an IoV environment, vehicles not only need to exchange information with roadside units through wireless communications, but also need to process navigation information, environmental perception data, and computing tasks generated by in-vehicle entertainment systems in real time during driving. The processing of these computing tasks requires an increasing amount of resources, so how to efficiently manage the resources of vehicle platoons has become an urgent problem to be solved.

[0003] At present, there are two main methods for the resource management of vehicle formations: one is based on vehicle formation control, which optimizes the vehicle spacing and fuel efficiency by optimizing the communication between vehicles and adjusting the vehicle parameters; the other is based on computing task processing, which improves the task processing efficiency by allocating computing tasks to roadside units. These methods optimize the management of vehicle formation resources from different perspectives.

[0004] However, existing technologies often separate vehicle formation control from computing task processing and optimize them separately, ignoring the mutual influence between the two. For example, although smaller vehicle spacing is conducive to improving road capacity, it will increase communication interference between vehicles and roadside units, affecting the processing efficiency of computing tasks. This makes it difficult for existing resource management methods to optimize the overall performance of vehicle formations. Summary of the invention

[0005] The present invention provides a resource management method and related equipment for joint vehicle formation control and task offloading, which can optimize the overall performance of the joint vehicle formation while ensuring the communication quality.

[0006] In a first aspect of the present invention, a resource management method for combined vehicle formation control and task offloading is provided, comprising: Acquiring driving information of a vehicle formation, wherein each vehicle in the vehicle formation establishes a communication connection with a roadside unit via an antenna; Based on the driving information, a multi-objective non-convex constrained optimization problem for the vehicle formation is defined, wherein the multi-objective non-convex constrained optimization problem is used to minimize the task processing delay and vehicle spacing of the vehicle formation under constraints; Solving the multi-objective non-convex constrained optimization problem to obtain optimization parameters; Resource management is performed on the vehicle formation based on the optimization parameters.

[0007] In a second aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a resource management method for a combined vehicle formation as described in any one of the above is implemented.

[0008] In a third aspect of the present invention, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the resource management method of the combined vehicle formation as described in any one of the above is implemented.

[0009] In a fourth aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the method for resource management of a combined vehicle formation as described in any one of the above is implemented.

[0010] In summary, one or more technical solutions provided in the present invention have at least the following technical effects or advantages: By obtaining the driving information of the vehicle formation and defining a multi-objective non-convex constrained optimization problem based on the driving information, the optimization problem comprehensively considers the task processing delay of the joint vehicle formation and the minimization goal of the vehicle spacing. The optimization parameters are obtained by solving the non-convex constrained optimization problem, and the resources of the joint vehicle formation are managed based on the optimization parameters, so that the vehicle formation control and the computing task processing are unified and optimized. Compared with the prior art, the present invention considers the vehicle spacing adjustment and the computing task processing in a unified way through the non-convex constrained optimization problem, avoids the problem of separating the two and optimizing them separately, and realizes the optimization of the overall performance of the joint vehicle formation while ensuring the communication quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 It is a flow chart of a resource management method for combined vehicle formation control and task offloading provided by an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of a scenario provided by an embodiment of the present invention.

[0014] Figure 3 It is a network framework diagram of a dual-time-scale SAC algorithm provided by an embodiment of the present invention.

[0015] Figure 4It is a small time scale algorithm convergence diagram provided by an embodiment of the present invention.

[0016] Figure 5 It is a large time scale algorithm convergence diagram provided by an embodiment of the present invention.

[0017] Figure 6 This is a delay comparison diagram of different algorithms provided by an embodiment of the present invention as the RSU computing power changes.

[0018] Figure 7 A delay comparison diagram of different algorithms provided in an embodiment of the present invention as the vehicle computing power changes.

[0019] Figure 8 This is a delay comparison diagram of different algorithms provided by an embodiment of the present invention as the number of vehicles changes.

[0020] Fig. 9 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please refer to Figure 1 , Figure 1 This is a flow chart of a method for managing resources of a vehicle formation provided by an embodiment of the present invention. The method can be implemented by a computer program, can be implemented by a single-chip microcomputer, or can be run on a resource management system for a vehicle formation based on a von Neumann system. The computer program can be integrated into an application or run as an independent tool application. Specifically, the method can include the following steps: S101. Acquire driving information of a vehicle formation, and each vehicle in the vehicle formation establishes a communication connection with a roadside unit via an antenna.

[0023] Among them, vehicle formation refers to organizing multiple vehicles in a coordinated manner through vehicle networking technology to form a closely coordinated team, and the vehicles share information and achieve coordinated control through wireless communication. In the embodiment of the present invention, it can be understood as a formation composed of multiple vehicles, and each vehicle travels in coordination on a horizontal road according to a predetermined formation and spacing.

[0024] Furthermore, vehicle platoons are used to improve road capacity, reduce traffic congestion and accident risks, and reduce pollutant emissions. Specifically, through real-time vehicle-to-vehicle communication and information sharing, each vehicle can automatically adjust its driving status, including speed synchronization, spacing adjustment, and path planning, to adapt to changes in road conditions and ensure the overall safety and stability of the fleet. In the present invention, the vehicle platoon also has the function of optimizing the allocation of computing tasks, that is, each vehicle can offload part of the computing tasks to the roadside unit (RSU) for processing based on its own computing power and communication conditions, thereby improving the computing efficiency of the entire system. This method of jointly optimizing formation spacing control and task offloading can achieve efficient use of computing resources while ensuring traffic efficiency.

[0025] Correspondingly, driving information refers to a set of dynamic parameters that describe the motion state and spatial position of a vehicle formation, including time-varying data such as vehicle position, speed, acceleration, and vehicle spacing. In the embodiment of the present invention, it can be understood as parameters such as the spatial position coordinates, motion velocity vector, real-time acceleration, spacing between adjacent vehicles, and delay characteristics when the vehicle responds to control instructions of each vehicle in the vehicle formation. These parameters can be obtained through the vehicle's own sensor system and shared in real time through the Internet of Vehicles.

[0026] Furthermore, driving information is used to monitor and predict the motion state of vehicle formations in real time, providing a decision-making basis for dual-time scale resource allocation schemes. By analyzing this driving information, the system can evaluate the stability of the formation, predict possible beam interference, and dynamically adjust parameters such as vehicle spacing, transmission power, and beam width. At the same time, driving information is also used to formulate computing task offloading strategies. The system can optimize the allocation of computing tasks based on the spatial distribution and motion state of the vehicles to ensure the stability of the communication link and the efficiency of task processing.

[0027] Based on the above embodiment, as an optional embodiment, the driving information may include vehicle spacing, vehicle expected acceleration, and spacing error. In S101, the step of obtaining the driving information of the vehicle formation may further include the following steps: S201. Obtain the vehicle distance between vehicles in a vehicle formation.

[0028] The vehicle spacing refers to the longitudinal distance between adjacent vehicles in the combined vehicle formation, reflecting the spatial distribution relationship between the vehicles in the formation. In the embodiment of the present invention, it can be understood as the actual distance between the front bumper of a vehicle and the rear bumper of the vehicle in front of it, which can be directly measured by the vehicle-mounted sensor system or calculated by the vehicle position information.

[0029] Furthermore, the vehicle spacing is used to evaluate the traffic efficiency and communication quality of the formation. Although smaller vehicle spacing can improve road capacity and fuel efficiency, it will increase the beam interference between vehicles and roadside units and affect task offloading performance; larger vehicle spacing can reduce beam interference and improve communication quality, but it will reduce road utilization. Therefore, real-time monitoring and adjustment of vehicle spacing is of great significance for balancing traffic efficiency and computing task processing efficiency, and is a key parameter for achieving joint vehicle formation control and computing task offloading optimization.

[0030] For example, please refer to Figure 2 , Figure 2 A schematic diagram of a scenario provided by an embodiment of the present invention. Figure 2 As shown, in a The platoon consists of 100 vehicles running on a horizontal road. PL is the 0th vehicle, PM is the A car, {1,..., }.set up and denotes the set of all vehicles and PMs respectively. Assume that the straight road is parallel to the three-dimensional Cartesian coordinate system. axis, and the origin is on the road. , 0, 0) indicates vehicle Coordinates of the midpoint of the rear bumper, [ ,0,0] indicates vehicle The velocity vector is The RSU is deployed on the roadside with coordinates (0, , ). Then the vehicle spacing can be defined as: in, Indicates the length of each vehicle.

[0031] S202: construct a vehicle dynamics model of the vehicle formation, where the vehicle dynamics model is used to characterize the motion states of the vehicles in the vehicle formation.

[0032] The vehicle dynamics model refers to a mathematical model that describes the motion law of each vehicle in the vehicle formation, which describes the relationship between state variables such as vehicle position, speed, acceleration, and the response characteristics of the vehicle to control commands. In the embodiment of the present invention, it can be understood as a simplified one-dimensional kinematic equation group, which takes into account the longitudinal motion characteristics of the vehicle on a horizontal road and the delay characteristics when the vehicle executes an acceleration or deceleration command.

[0033] Furthermore, the vehicle dynamics model is used to accurately describe and predict the motion behavior of vehicle formations, evaluate the stability of the formation, and provide a theoretical basis for the control decision of the system. Through the vehicle dynamics model system, the control strategy of vehicle spacing can be optimized on a large time scale, and the changes in the communication environment can be predicted on a small time scale, thereby achieving dynamic optimization of transmission power, beam width, and task offloading strategy, and ultimately achieving the goal of balancing traffic efficiency and computing efficiency.

[0034] For example, when the vehicle is When the axis moves, one-dimensional kinematics can be used to describe the vehicle dynamics model as follows: in, represents the constant duration of the actuation lag in response to any command to accelerate or decelerate, Indicates vehicle The required acceleration, , and Respectively represent vehicles position, velocity and acceleration.

[0035] S203: Determine the expected acceleration and spacing error of the vehicles in the vehicle formation based on the vehicle dynamics model and the vehicle spacing.

[0036] Among them, the expected acceleration of the vehicle refers to the ideal acceleration value that the vehicle needs to execute in order to maintain the stable operation of the formation and maintain the target spacing. This value is calculated by comprehensively considering the motion state of the front and rear vehicles, the speed difference, and the spacing adjustment requirements. The spacing error refers to the deviation between the actual vehicle spacing and the preset target spacing, which is used to measure the accuracy of the formation control. In the embodiment of the present invention, it can be understood that: the expected acceleration of the vehicle is a comprehensive control quantity, which reflects the acceleration or deceleration behavior that the vehicle needs to take according to the overall state of the formation; the spacing error is a real-time monitoring quantity, which reflects the control effect of the vehicle spacing.

[0037] Furthermore, the spacing error is used to achieve precise control and status evaluation of the vehicle formation. By continuously monitoring this parameter, the system can adjust the vehicle's acceleration and other motion states in a timely manner to ensure the stability and safety of the formation. At the same time, this information also provides important decision-making basis for the dual-time scale resource allocation scheme, which helps the system optimize the allocation of communication resources and task processing efficiency while ensuring the formation control effect.

[0038] Exemplarily, the spacing error can be defined as: The expected acceleration of the vehicle is: In the formula, Indicates vehicle and vehicles 1 speed difference between,hyperparameters is the weight of the acceleration and speed of the front and rear vehicles. is the damping ratio, and the critical damping can be taken as 1. Parameters is the control gain used in the longitudinal controller. The desired acceleration of PL Known by the controller, the desired acceleration can then be determined PM by PM.

[0039] Based on the above embodiment, as an optional embodiment, the driving information may include a signal transmission rate. In S101, the step of obtaining the driving information of the vehicle formation may further include the following steps: S301. Construct a millimeter wave communication model of a vehicle platoon, where the millimeter wave communication model is used to characterize the communication interaction process between the vehicle platoon and the roadside unit.

[0040] The millimeter wave communication model refers to a mathematical model that describes the signal transmission characteristics between each vehicle in the vehicle formation and the roadside unit. The model describes various factors affecting the millimeter wave signal during the propagation process, including line-of-sight transmission, non-line-of-sight transmission, path loss, beam interference and other characteristics. In the embodiment of the present invention, it can be understood as a communication system description model that comprehensively considers vehicle position, antenna configuration, and channel characteristics. The model reflects the transmission law and performance characteristics of millimeter wave signals in a vehicle formation environment.

[0041] Furthermore, the millimeter wave communication model is used to evaluate and predict the communication quality between vehicles and roadside units, providing a basis for the system's resource allocation decisions. Through the millimeter wave communication model, the system can analyze the impact of changes in vehicle spacing on communication performance and predict possible beam interference, thereby guiding the dynamic adjustment of transmission power and beam width to ensure the reliability of the communication link.

[0042] For example, assume that each vehicle is equipped with one antenna. At the same time, the vehicles in the formation are served by RSUs equipped with M antenna uniform linear arrays (ULA). The length of each time slot is The index set of time slots is represented as ∈ in Indicates the total number of time slots, time slot Inside From vehicle The channel coefficient vector to RSU can be modeled as: in, and Respectively represent time slots Vehicles during the period The complex channel gain and normalized direction of the LOS path, and Respectively represent time slots Vehicles during the period No. The complex channel gain and normalized direction of the NLOS path. The array steering vector of the ULA can be defined as: in Represents the imaginary part. Normalized direction is normalized by the physical AoA The obtained . and are the antenna spacing and carrier wavelength at the RSU respectively. Physical AoA with RSU and vehicle The relationship between the positions can be defined as: A location-based beamforming method is used for V2I communication. vehicle The estimated position within the initial row of is defined as: in It is a vehicle In time slot The random position error at , following the standard deviation (STD) , The unknown probability distribution of . Based on the vehicle's location information, calculate the vehicle In time slot The estimated direction of the LOS path within is: The receiving end encoding vector is is the size of the coefficient vector at the receiving end, .

[0043] S302: Obtain a signal-to-interference-and-noise ratio between a roadside unit and a vehicle formation based on a millimeter wave communication model.

[0044] Among them, the signal-to-interference-noise ratio refers to the ratio of the useful signal power to the sum of the interference signal power and the noise power at the receiving end of the roadside unit, which reflects the quality level of the communication link. In the embodiment of the present invention, it can be understood as a signal quality indicator when the roadside unit receives a vehicle signal, where the useful signal comes from the transmission of the target vehicle, and the interference signal mainly comes from the transmission of other vehicles in the formation, and also includes the influence of environmental noise.

[0045] Furthermore, the signal-to-interference-to-noise ratio is used to evaluate the quality of the communication link between the vehicle and the roadside unit, providing an important decision-making basis for system resource allocation. By monitoring the changes in the signal-to-interference-to-noise ratio, the system can promptly detect problems with reduced communication quality and take corresponding compensation measures, such as adjusting the target vehicle spacing, transmission power or beam width.

[0046] For example, the SINR of the vehicle signal received by the RSU can be defined as: S303: Determine a signal transmission rate of the vehicle formation based on the signal to interference and noise ratio.

[0047] The signal transmission rate refers to the amount of data that can be transmitted between the vehicle and the roadside unit per unit time, and the rate reflects the transmission capacity of the communication link. In the embodiment of the present invention, it can be understood as the maximum reliable data transmission rate that the communication system can achieve under a given signal-to-interference-to-noise ratio, and its size directly affects the efficiency of computing task offloading.

[0048] For example, the transmission rate can be defined as: Based on the above embodiment, as an optional embodiment, the driving information may include a first calculation delay for the vehicle formation to perform a local task and a second calculation delay for the roadside unit to process an unloading task. In S101, the step of obtaining the driving information of the joint vehicle formation may further include the following steps: S401, constructing a computational offloading model for a vehicle formation, where the computational offloading model is used to characterize the vehicle task processing process of the vehicle formation and the roadside unit, where the vehicle tasks include local tasks of the vehicle formation and offloading tasks transmitted to the roadside unit.

[0049] The computational offloading model refers to a mathematical model that describes the computational task allocation and processing process in a vehicle formation. The computational offloading model characterizes the distribution, transmission, and processing characteristics of tasks between the on-board computing unit and the roadside unit. In the embodiment of the present invention, it can be understood as a system description model that comprehensively considers the computational resource allocation, task transmission overhead, and processing delay. The model reflects the resource consumption and performance characteristics of the two processing methods of local task processing and task offloading.

[0050] Furthermore, the computational offloading model is used to evaluate and optimize the system's task processing performance and provide a theoretical basis for task allocation decisions. Through this model, the system can analyze the resource consumption and processing efficiency of different task allocation schemes, thereby guiding the optimal allocation of tasks and achieving efficient utilization of computing resources.

[0051] For example, , Represents the maximum CPU cycle frequency of each vehicle and each RSU respectively. At the beginning of each time slot, the vehicle There is a periodic computationally intensive task that needs to be processed in time, modeled as Here, let Indicates the amount of original task data. (CPU cycle / bit) represents processing density, that is, the number of CPU cycles required to process one unit of data. (ms) represents the task delay requirement. The vehicle task is divided into two parts, and the proportion of local computing tasks to the total task volume is The remaining part is calculated in RSU, and the first-come-first-served principle is adopted. Tasks with low transmission delay are processed first, and after processing, their computing resources are allocated to unprocessed tasks. Definition is the computation time required for the remaining uncomputed tasks of vehicle i in the previous time slot, and is defined as It is the computing time required for the remaining uncomputed tasks of the RSU in the previous time slot.

[0052] S402: Based on the computational offloading model, obtain a first computational delay for the vehicle formation to perform a local task and a second computational delay for the roadside unit to process an offloading task.

[0053] Among them, the first computing delay and the second computing delay refer to the time overhead required to complete the computing task on different processing units. Among them, the first computing delay refers to the time required for the task to be processed locally on the on-board computing unit, and the second computing delay refers to the total time required for the task to be transmitted to the roadside unit and completed, including the data transmission delay and the actual processing delay of the roadside unit. In the embodiment of the present invention, it can be understood as two key time indicators for measuring the efficiency of task processing, which respectively reflect the time performance of local processing and remote processing.

[0054] Furthermore, the first computing delay and the second computing delay are used to evaluate the time efficiency of different task processing solutions, providing an important basis for the system to make task allocation decisions. By comparing these two delay indicators, the system can select a better task processing solution and achieve efficient use of computing resources.

[0055] Exemplarily, the first computing delay of the local computing task can be expressed as: The task transmission delay of the task at the RSU can be expressed as: The task calculation delay in RSU can be expressed as: Furthermore, the waiting delay can be expressed as Then the second computation delay can be expressed as .

[0056] The vehicle task calculation delay is S102. Define a multi-objective non-convex constrained optimization problem for joint vehicle formation control and computational offloading based on driving information. The multi-objective non-convex constrained optimization problem is used to minimize the task processing delay and vehicle spacing of the vehicle formation under constraints.

[0057] Among them, the multi-objective non-convex constrained optimization problem refers to a type of mathematical optimization problem that seeks the optimal solution on a non-convex feasible domain, and the objective function or constraint of the problem contains a non-convex function, so that the solution space of the problem presents a non-convex characteristic. In the embodiment of the present invention, it can be understood as a multi-objective optimization problem that comprehensively considers the task processing delay and vehicle spacing, which includes nonlinear constraints such as inter-beam interference, formation stability, and task offloading delay.

[0058] Furthermore, the multi-objective non-convex constrained optimization problem is used to find the optimal resource allocation strategy under the premise of satisfying various system constraints. By uniformly modeling the vehicle formation control and task offloading problems as a multi-objective non-convex constrained optimization problem, the system can seek a balance between the two objectives of vehicle spacing and task processing delay to achieve the optimization of overall performance. Specifically, smaller vehicle spacing is conducive to improving road capacity and fuel efficiency, but it will aggravate beam interference between vehicles, affecting communication quality and task offloading efficiency; while larger vehicle spacing can reduce beam interference and improve communication quality, but it will reduce road utilization. At the same time, the allocation ratio of tasks between local processing and offloading processing will also affect the overall performance of the system.

[0059] Constraints refer to a series of mathematical expressions that limit the system's operating state and control decisions. These expressions reflect the rules that the system must follow in terms of physical characteristics, safety requirements, and performance indicators. In the embodiments of the present invention, it can be understood as a restriction on the value range and mutual relationship of various parameters in the process of vehicle formation control and task unloading, and the optimal balance point between the constraints.

[0060] In a feasible implementation, the constraint condition includes at least one of the following conditions: Vehicle spacing constraints and queue stability constraints for vehicle platooning.

[0061] Beam width range constraints, transmission power constraints, variable range constraints for task offloading, and delay constraints for task offloading during communication interaction between vehicle platoons and roadside units.

[0062] Exemplarily, the above non-convex constrained optimization problem can be expressed as follows: Where C1 is the beam width range constraint, C2 is the transmit power constraint, C3 is the offloading decision variable range constraint, C4 is the task offloading delay constraint, C5 is the vehicle spacing constraint, and C6 is the queue stability constraint. Since the expressions of average spacing and task delay are difficult to represent analytically, and the PL trajectory and traffic environment are uncertain, theoretical modeling is also challenging. Finally, there are significant differences in the temporal characteristics of the communication system and the on-board system. Specifically, the beamforming design can be implemented at the millisecond level. However, due to the physical limitations of the on-board system, the desired vehicle spacing is delayed and is implemented at the second level. To this end, a dual-time scale reinforcement learning framework is proposed to jointly optimize the beamforming design, spacing strategy, and task offloading strategy in vehicle formations.

[0063] S103. Solve the multi-objective non-convex constrained optimization problem and obtain optimization parameters.

[0064] Specifically, the established multi-objective non-convex constrained optimization problem is solved by a dual-time-scale reinforcement learning algorithm to obtain the optimal optimization parameters required for system operation. In the embodiment of the present invention, it can be understood that the deep reinforcement learning method is used to learn and optimize the vehicle spacing control strategy and the transmission power, beam width and task offloading decision on a large time scale and a small time scale, respectively, so as to obtain the optimal optimization parameter configuration that meets various constraints.

[0065] Furthermore, the reason for using a dual-time-scale reinforcement learning algorithm to solve the multi-objective non-convex constrained optimization problem is that: first, since the vehicle control system and the communication system have significantly different time characteristics, the traditional single-time-scale optimization method is difficult to effectively handle this difference; second, the state space and action space involved in the problem are large, and there are multiple nonlinear constraints, which are difficult to be solved directly by traditional optimization algorithms. By introducing a dual-time-scale deep reinforcement learning algorithm based on SAC (Soft Actor-Critic), the system can gradually learn the optimal control strategy based on the consideration of time scale differences. Specifically, on a small time scale, the algorithm quickly adjusts the transmit power, beam width, and task offloading decisions based on the currently observed state information, including channel conditions, computational load, etc., to adapt to the real-time changes in the communication environment; on a large time scale, the algorithm performs long-term planning and optimization of vehicle spacing based on the accumulated system performance evaluation results to balance traffic efficiency and communication efficiency.

[0066] Based on the above embodiment, as an optional embodiment, in S103: solving the multi-objective non-convex constrained optimization problem to obtain the optimization parameters may further include the following steps: S501. Define the decision interval corresponding to the multi-objective non-convex constrained optimization problem as NThe first time slot and K The second time slot, N The sum of the first time slots is equal to K The sum of the second time slots, N , K is a positive integer, and N Greater than K .

[0067] Among them, the decision interval refers to the complete time period for the system to make control decisions, which includes the execution time of all control operations of the system. In the embodiment of the present invention, it can be understood as a complete operation cycle including multiple control cycles of different time scales, during which the system needs to complete all decision tasks from communication control to vehicle formation control. The decision interval is used to provide a unified time reference framework for the system, and by reasonably dividing the control cycles of different scales, it ensures that the various control modules of the system can operate in a coordinated manner to achieve overall performance optimization.

[0068] Furthermore, the first time slot refers to the basic time unit for the system to make rapid control decisions, and the time unit corresponds to the response characteristics of the communication system and the task processing system. In the embodiment of the present invention, it can be understood as the minimum time interval for the system to adjust the transmission power, beam width and task offloading decision, and its length needs to meet the real-time requirements of the communication system. The first time slot is used to ensure that the system can respond to changes in the communication environment in a timely manner, maintain communication quality and computing efficiency through frequent parameter adjustments, and realize dynamic optimization allocation of resources.

[0069] Correspondingly, the second time slot refers to the basic time unit for the system to make stable control decisions, which corresponds to the dynamic characteristics of the vehicle control system. In the embodiment of the present invention, it can be understood as the minimum time interval for the system to adjust the target vehicle spacing, and its length needs to consider the physical response characteristics of the vehicle and the formation stability requirements. The second time slot is used to ensure the stability of vehicle formation control, and to avoid the impact of frequent adjustments on formation stability through a longer control cycle. At the same time, it accumulates sufficient performance evaluation information for the system to support better control decisions.

[0070] Exemplarily, the decision interval for controlling the target vehicle spacing includes multiple time slots, and the time length of each time slot is , time slot index is ,in , represents the number of time slots in the large time scale. Assume > , = N ,definition In the At the beginning of the time slot, the agent obtains the state , and according to its strategy ) Take Action Large time scale control, time slot At the end, the agent gets rewarded , the environment is transferred to the state . In each small time slot. At the beginning, the agent can obtain the state , and according to the strategy Take Action Perform small time-scale power, beam width, and unloading decision control. Then, in the time slot At the end, update the next state , get rewarded Based on the information collected at small time scales, the small time scale strategy can be updated .

[0071] S502, convert the multi-objective non-convex constrained optimization problem into N The first Markov process under the first time slot, and K The second Markov process in the second time slot.

[0072] Among them, Markov Decision Processes (MDP) is a mathematical framework for describing sequential decision problems with random properties. The MDP model describes the interaction between the agent and the environment. At the beginning, the agent obtains the state , and then according to the strategy Take Action ,in and Represents the set of states and actions respectively. At the end, the status Transition to the next state , the environment returns a reward ,in is the reward function. The cumulative discount reward is defined as ,in is the discount factor for future reward weights.

[0073] Among them, the first Markov process refers to a mathematical model that describes the system state transition and decision-making process on a small time scale. The model has the Markov property, that is, the system state at the next moment depends only on the current state and action, and has nothing to do with the historical state. In the embodiment of the present invention, it can be understood as a dynamic decision-making process that describes the communication system and the task processing system in each first time slot, including elements such as system state, action space, state transition probability, and reward function. The system state reflects the rapidly changing environmental information such as vehicle position error, channel state, and computing task characteristics, and the action space includes the adjustment decision of transmission power, beam width, and task offloading ratio. The first Markov process is used to establish a mathematical description of the dynamic characteristics of the system on a small time scale. Through the design of state transition and instant reward, it guides the system to learn the optimal communication control and task offloading strategy to achieve real-time optimization allocation of resources.

[0074] Correspondingly, the second Markov process refers to a mathematical model that describes the state transition and decision-making process of the system on a large time scale. The model also has the Markov property, but its state transition cycle is longer. In the embodiment of the present invention, it can be understood as a description of the dynamic decision-making process of the vehicle formation control system in each second time slot, which includes elements such as system state, action space, state transition probability, and reward function. The system state reflects slowly changing physical quantities such as vehicle speed, acceleration, and average spacing, and the action space is the adjustment decision of the vehicle spacing. The second Markov process is used to establish a mathematical description of the dynamic characteristics of the system on a large time scale. Through the comprehensive evaluation of the system performance in multiple first time slot cycles, it guides the optimal control of the vehicle spacing and achieves a long-term balance between traffic efficiency and communication efficiency.

[0075] Based on the above embodiments, as an optional embodiment, the state of the first Markov decision process is defined as key information reflecting the communication environment and task offloading, the action is defined as the decision corresponding to the transmit power, beam width and task offloading ratio, and the reward function includes functions corresponding to the task processing delay and the task delay constraint.

[0076] Among them, the state of the first Markov decision process refers to a set of key variables that describe the operating environment of the system on a small time scale. These variables can fully reflect the current communication conditions and computing load of the system. In the embodiment of the present invention, it can be understood as a multidimensional vector containing information such as the channel state from the vehicle to the roadside unit, the vehicle position estimation error, the characteristic parameters of the computing task, and the current computing load. The channel state reflects the communication quality, the position error affects the beam alignment accuracy, and the task characteristics and computing load determine the task processing efficiency. The state definition is used to provide a decision-making basis for the system. By monitoring these state variables in real time, the system can perceive environmental changes in a timely manner and provide necessary information support for communication control and task offloading decisions.

[0077] Among them, the action of the first Markov decision process refers to the set of decisions that the system can take on a small time scale, and these decisions directly affect the communication quality and task processing efficiency. In the embodiment of the present invention, it can be understood as a three-dimensional vector containing the transmit power adjustment amount, the beam width adjustment amount, and the task offloading ratio, where the transmit power determines the signal transmission quality, the beam width affects the signal coverage and interference level, and the task offloading ratio determines the task allocation ratio of local computing and remote computing. The action definition is used to clarify the control space of the system. Through the combination of these control variables, the system can flexibly adjust the resource allocation strategy to achieve dynamic optimization of communication efficiency and computing efficiency.

[0078] Among them, the reward function of the first Markov decision process refers to a numerical indicator for evaluating the control decision effect of the system on a small time scale, which comprehensively considers the performance goals and constraint requirements of the system. In the embodiment of the present invention, it can be understood as a scalar function that simultaneously reflects the task processing delay and the degree of satisfaction of the delay constraint. When the system can complete the task processing within the constraint delay, it will obtain a positive reward, and if it exceeds the constraint delay, it will be punished. The reward function is used to guide the system to learn the optimal control strategy. Through the design of instant rewards, the system can learn the control strategy that minimizes the processing delay while ensuring the real-time performance of task processing.

[0079] Exemplarily, for the state of the first Markov decision process: in each time slot, the agent only makes decisions on the transmission power, beam width, and task offloading ratio of the concurrent V2I transmission between the vehicle and the RSU. The state at the start time is the time slot The difference in AoA, uncertainty in position error, mission characteristics, and the computational load of the current vehicle and RSU are crucial to determining the transmit power, beam width, and mission offloading ratio. Therefore, the state at a small time scale is: in, Indicates time slot The estimated normalized AoA of the vehicle with estimation error at the beginning, Indicates time slot The STD of the vehicle position estimation error at the start time. represents the original task data volume, processing density and task delay requirements, It is the calculation time required for the remaining uncalculated tasks of the vehicle in the previous time slot and the calculation time required for the remaining uncalculated tasks of the RSU in the previous time slot.

[0080] For the actions of the first Markov decision process: The actions at small time scales include transmit power, beam width, and task offloading decisions. Therefore, the action is defined as: For the reward of the first Markov decision process: The purpose of controlling the transmit power, beam width and task offloading decision is to ensure that the task offloading delay of each vehicle is as small as possible while satisfying the constraints. Since the transmit power, beam width and task offloading decision have no direct impact on the vehicle spacing and fuel efficiency, and no decision is made on the target spacing on a small time scale, the reward can only consider the task offloading delay given the current target vehicle spacing. Define the reward as: in, represents the total delay of all vehicles performing computing tasks locally, is the penalty function, when hour, , otherwise you will be punished.

[0081] Based on the above embodiment, as an optional embodiment, the state of the second Markov decision process is defined as the physical state of the vehicle and the overall condition of the vehicle formation, the action is defined as the decision corresponding to the vehicle spacing, and the reward function includes minimization of vehicle spacing, minimization of task offloading delay, and functions corresponding to safety constraints.

[0082] Among them, the state of the second Markov decision process refers to a set of key variables that describe the operating status of the vehicle formation of the system on a large time scale. These variables reflect the dynamic characteristics of the vehicle and the overall performance of the formation. In the embodiment of the present invention, it can be understood as a multidimensional vector containing information such as vehicle speed, acceleration, estimated heading angle, standard deviation of vehicle position error, current vehicle spacing, and average task processing delay of the previous decision cycle, among which the dynamic parameters reflect the motion state of the vehicle, the position error reflects the positioning accuracy, and the spacing and delay information reflect the overall performance of the system. The state definition is used to fully grasp the operating status of the vehicle formation. Through continuous monitoring of these state variables, the system can accurately evaluate the stability of the formation and the efficiency of task processing, and provide a decision-making basis for the optimal control of vehicle spacing.

[0083] Among them, the action of the second Markov decision process refers to the set of control decisions that the system makes to adjust the vehicle spacing on a large time scale, which directly affects the traffic efficiency and communication quality of the formation. In the embodiment of the present invention, it can be understood as a vector representing the desired vehicle spacing, which defines the target distance that should be maintained between adjacent vehicles in the formation. The action definition is used to standardize the spatial structure of the vehicle formation. By reasonably setting the vehicle spacing, the system can ensure communication quality and driving safety while improving the road capacity, and achieve a balance between traffic efficiency and communication efficiency.

[0084] Among them, the reward function of the second Markov decision process refers to a comprehensive indicator for evaluating the control decision effect of the system on a large time scale, which balances multiple performance objectives. In the embodiment of the present invention, it can be understood as a scalar function that simultaneously considers the average vehicle spacing, task processing delay, and the degree of satisfaction of safety constraints. When the system can achieve smaller vehicle spacing and task processing delay while ensuring safety, it will obtain a higher reward. The reward function is used to guide the system to learn the optimal spacing control strategy. Through the accumulation of long-term rewards, the system can gradually find the optimal balance between safety, traffic efficiency, and communication efficiency.

[0085] For example, for the state of the second Markov decision process: the agent's decision on vehicle spacing needs to balance traffic capacity and communication efficiency to minimize vehicle spacing and computational unloading delay. Therefore, information related to communication and vehicle dynamics is needed to determine the spacing. The initial state is , then the state of the second Markov decision process can be expressed as: in, Estimate AoA for the vehicle, is the vehicle position error STD, , , are the vehicle speed, acceleration and vehicle distance, respectively. is the average unloading delay in the previous spacing decision cycle.

[0086] For the actions of the second Markov decision process: at a large time scale, the agent controls the spacing within the queue, and the action is expressed as: Indicates vehicle and vehicles The acceleration of each vehicle can be calculated based on the expected spacing.

[0087] Rewards for the second Markov decision process: The goal of spacing control is to minimize the vehicle spacing and computational unloading delay. Therefore, when the spacing constraints are met, the agent should receive positive rewards for obtaining a shorter vehicle spacing or a smaller task unloading delay. Based on the above principles, the reward design for spacing control is as follows: S503, solving the first Markov decision process to obtain the transmission power, beam width and task offloading decision parameters of the vehicle formation; solving the second Markov decision process to obtain the target vehicle spacing decision parameters of the vehicle formation.

[0088] Specifically, in the embodiment of the present invention, a deep reinforcement learning method based on the SAC algorithm is adopted to learn the optimal control strategy of transmission power, beam width and task offloading ratio through continuous interaction with the environment, so as to achieve performance optimization of the system on a small time scale.

[0089] Furthermore, the reason for choosing the SAC algorithm for solving is that the state space and action space of the system are both continuous, and traditional discrete reinforcement learning methods are difficult to apply directly; secondly, the SAC algorithm achieves a good balance between exploration and utilization by introducing the maximum entropy framework, which helps to find a more robust control strategy; finally, the algorithm adopts a random strategy based on temperature parameters, which can better cope with environmental uncertainty. In the specific implementation process, the system first constructs a deep neural network structure containing a policy network and a value function network. The policy network is used to generate control decisions, and the value function network is used to evaluate the long-term value of decisions.

[0090] The experience replay mechanism stores the transformation samples generated by the interaction between the system and the environment, and uses these samples to update the network parameters. At the beginning of each first time slot, the system generates control decisions on the transmission power, beam width, and task offloading ratio through the policy network based on the currently observed state information; at the end of the time slot, the experience replay pool is updated and the network parameters are optimized based on the obtained reward signal and new state observations.

[0091] For example, Figure 3 As shown, Figure 3 A network framework diagram of a dual-time-scale SAC algorithm provided by an embodiment of the present invention. The SAC algorithm is an advanced off-policy deep reinforcement learning algorithm based on the maximum entropy framework. In this framework, the agent aims to maximize both expected return and entropy. Entropy maximization improves the stability and exploratory nature of the learning process and is robust to model and estimation errors. The goal of the traditional actor-critic algorithm is to find a strategy that maximizes expected return. , the SAC algorithm adds an entropy term to the objective. The optimal random strategy that maximizes expected return and entropy is expressed as ; where α>0 is the temperature parameter that determines the importance of the entropy term relative to the reward, yes For RL with entropy maximization, the state action value function is changed to a soft Q function defined as follows: ; In order to handle large-scale continuous action spaces, in the SAC algorithm, both the soft Q function (critic) and the policy (actor) are approximated by functions that can be optimized using stochastic gradient descent. Deep neural networks (DNNs) are used to approximate the soft Q function and the policy. The parameterized soft Q function and the parameterized policy are respectively and DNNs approximation, that is and The policy network is used to determine the mean and variance of each policy. During training, samples are taken from a Gaussian distribution, whose mean and variance are determined by the policy network. A pruned dual-Q strategy is used to reduce Q-value overestimation and improve learning performance. Two Q functions are used, namely and , the smaller value is used to calculate the target value for updating the network.

[0092] For the SAC algorithm that controls the transmit power, beamwidth, and task offloading decisions on a small time scale, in the time slot Initial, state-based , the agent chooses an action as follows: ; in represents a parameterized stochastic policy for action control at small time scales. Since the soft Q function is approximated by a differentiable parameterized neural network, the reparameterization trick can be used to reduce the variance of the estimator. By reparameterizing the policy using a neural network transformation, the actions are based on Take samples, is the input noise vector sampled from a spherical Gaussian distribution. At the end, the tuple is converted Store to buffer In order to reduce the correlation between samples, uniformly sample a Mini-batches of transformed tuples To update the parameters of DNNs to make them close to the policy and value functions. The parameters of the two critic networks are updated by minimizing the Bellman error loss function. ; In the actor-critic framework that considers entropy maximization, the actor should take a soft state value function that maximizes the expected return and entropy. Actions, parameters of the actor network Updated as follows: ; For the SAC algorithm for spacing control under large time scales, at the beginning of the kth time slot, the agent observes the state , and according to the strategy )Select an action ,in ) represents the parameterized spacing control strategy on a large time scale, represents the parameters. The strategy is reparameterized as , is the input noise vector. At the end of time slot k, the agent observes the state The evolved state , and get the execution action Remuneration , will convert the tuple Store to buffer In, from uniformly sample a Mini-batches of transformed tuples To update the parameters of DNNs to approximate the policy and value functions. Two value functions for spacing control are learned on a large time scale, namely and , update the parameters of the two critic networks by minimizing the Bellman error loss function: ; Target value It is obtained by the minimum of the two Q values.

[0093] ; Two target critic networks for distance control at large time scales. Represents the parameters of the two target critic networks. The parameters of the actor network Updated as follows: ; In order to ensure the stability of the learning process, the target evaluation network is updated slowly in <<1 can ensure that the target network is updated slowly.

[0094] S104. Perform resource management on the vehicle formation based on the optimization parameters.

[0095] Specifically, the system implements resource management in a hierarchical decision-making manner, which can effectively handle the time scale differences of different needs. At the small time scale level, the system first obtains the current communication environment status, including channel conditions, computing task characteristics and other information, and then adjusts the transmission power and beam width of each vehicle in real time according to the learned decision strategy to optimize the communication quality between the vehicle and the roadside unit; at the same time, based on the current computing load status, the ratio of local processing to remote processing of tasks is dynamically adjusted to ensure efficient use of computing resources. At the large time scale level, the system comprehensively evaluates the system performance in multiple small time scale cycles, including indicators such as communication quality, task processing efficiency and traffic efficiency, and adjusts the spacing configuration of the vehicle formation accordingly to achieve a balance between traffic efficiency and communication efficiency while ensuring safety.

[0096] Based on the above embodiment, the effect of the above method is verified and explained through simulation results.

[0097] For V2I communication between vehicles and RSU, the carrier frequency is set to 28 GHz. The antenna spacing is set to half wavelength, that is, The path loss model of V2I communication is expressed as , The variance is Lognormal shading of . Indicates the distance between the transmitter and the receiver (meters). For the LOS path, the parameter is set to =61.4, =2, =5.8. For the NLOS path, the parameter is set to =72.0, =2.92, = 8.7. For beam width control, the beam width is controlled by changing the amplitude distribution of the precoding coefficients. For the considered M-antenna ULA with half-wavelength spacing, the spatial null-to-null beam width is The relationship between the number of antennas is =4 / M. Therefore, a wider beamwidth can be obtained by reducing the number of active antennas, given a desired beamwidth ,when When the amplitude vector No. The elements are , otherwise 0.

[0098] In order to verify the effectiveness of the proposed dual-time-scale deep reinforcement learning algorithm, the following benchmark algorithms are used: Benchmark Algorithm 1): Using the dual-time-scale SAC algorithm, the communication rate is maximized and the average distance is minimized, and the tasks are only calculated locally. Benchmark Algorithm 2): Using the dual-time-scale SAC algorithm, the communication rate is maximized and the average distance is minimized, and the task offloading ratio is randomly assigned. Benchmark Algorithm 3): Using the dual-time-scale SAC algorithm, the communication rate is maximized and the average distance is minimized, and the tasks are only calculated at the RSU.

[0099] Please refer to Figure 4 and Figure 5 , Figure 4 A small time scale convergence diagram provided by an embodiment of the present invention, Figure 5 A large time scale convergence diagram is provided for an embodiment of the present invention. Figure 4 and Figure 5 It can be seen that the proposed algorithm can converge quickly on both time scales.

[0100] Please refer to Figure 6 , Figure 6 A delay comparison diagram of different algorithms provided by the embodiment of the present invention as the RSU computing power changes, Figure 6 It can be seen that as the computing power of RSU increases, the total latency of baseline 1 does not change significantly with the change of RSU computing power, because RSU computing power has no effect on the situation where the task is only calculated locally in the vehicle. The total task calculation latency of the proposed scheme and the scheme where the task is only calculated at RSU is significantly reduced. When the computing power of RSU reaches 9Ghz, the performance of the scheme where the task is only calculated at RSU is close to that of the proposed scheme. Figure 7 This is a delay comparison diagram of different algorithms provided by an embodiment of the present invention as the vehicle computing power changes. The vehicle computing power has no effect on the task when it is only calculated at the RSU, so the total delay of baseline 3 does not change significantly with the change of vehicle computing power. The total calculation delay of the remaining solutions decreases as the vehicle computing power increases.

[0101] Please refer to Figure 8 , Figure 8 A delay comparison diagram of different algorithms provided by the embodiment of the present invention as the number of vehicles changes. As the number of vehicles increases, the total delay of task calculation of all schemes increases. When the number of vehicles increases, the RSU calculation load increases, so the performance gap between baseline 3 and baseline 1 decreases as the number of vehicles increases. It can be seen that the proposed invention scheme can reasonably allocate resources under different parameter configurations and significantly reduce the total delay of task calculation.

[0102] Please refer to Fig. 9 , Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the resource management method of the combined vehicle formation control and computation offloading.

[0103] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0104] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the resource management method for combined vehicle formation control and task offloading provided by the above-mentioned methods.

[0105] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the resource management method for combined vehicle formation control and task offloading provided by the above-mentioned methods.

[0106] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0107] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A resource management method for combined vehicle formation control and task offloading, characterized in that: include: Acquiring driving information of a vehicle formation, wherein each vehicle in the vehicle formation establishes a communication connection with a roadside unit via an antenna; Based on the driving information, a multi-objective non-convex constrained optimization problem for the vehicle formation is defined, wherein the multi-objective non-convex constrained optimization problem is used to minimize the task processing delay and vehicle spacing of the vehicle formation under constraints; Solving the multi-objective non-convex constrained optimization problem to obtain optimization parameters; Resource management is performed on the vehicle formation based on the optimization parameters.

2. The resource management method for combined vehicle formation control and task offloading according to claim 1, characterized in that: The obtaining of driving information of the vehicle formation includes: Obtain the vehicle spacing between vehicles in the vehicle formation; Constructing a vehicle dynamics model of the vehicle formation, wherein the vehicle dynamics model is used to characterize the motion state of the vehicles in the vehicle formation; Determining expected acceleration and spacing error of vehicles in the vehicle formation based on the vehicle dynamics model and the vehicle spacing; The vehicle distance, the vehicle expected acceleration, and the distance error are used as driving information of the vehicle formation.

3. The resource management method for combined vehicle formation control and task offloading according to claim 1, characterized in that: The obtaining of driving information of the vehicle formation includes: Constructing a millimeter wave communication model of a vehicle formation, wherein the millimeter wave communication model is used to characterize a communication interaction process between the vehicle formation and the roadside unit; Acquire a signal-to-interference-and-noise ratio between the roadside unit and the vehicle formation based on the millimeter wave communication model; Determining a signal transmission rate of the vehicle formation based on the signal to interference and noise ratio; The signal transmission rate is used as driving information of the vehicle formation.

4. The resource management method for combined vehicle formation control and task offloading according to claim 1, characterized in that: The obtaining of driving information of the vehicle formation includes: Constructing a computational offloading model of a vehicle formation, wherein the computational offloading model is used to characterize a vehicle task processing process of the vehicle formation and the roadside unit, wherein the vehicle task includes a local task of the vehicle formation and an offloading task transmitted to the roadside unit; Based on the computational offloading model, obtaining a first computational delay for the vehicle formation to perform a local task and a second computational delay for the roadside unit to process the offloading task; The first calculation delay and the second calculation delay are used as driving information.

5. The resource management method for combined vehicle formation control and task offloading according to claim 1, characterized in that: The constraint condition includes at least one of the following conditions: The vehicle spacing constraint and the queue stability constraint of the vehicle formation; The beam width range constraints, transmission power constraints, variable range constraints for task offloading, and delay constraints for task offloading during the communication interaction between the vehicle formation and the roadside unit.

6. The resource management method for combined vehicle formation control and task offloading according to claim 1, characterized in that: The step of solving the multi-objective non-convex constrained optimization problem to obtain optimization parameters includes: The decision interval corresponding to the multi-objective non-convex constrained optimization problem is defined as N The first time slot and K The second time slot, N The sum of the first time slots is equal to K The sum of the second time slots, N , K is a positive integer, and N Greater than K ; The multi-objective non-convex constrained optimization problem is transformed into N The first Markov process under the first time slot, and K The second Markov process under the second time slot; Solving the first Markov decision process to obtain parameters corresponding to the transmission power, beam width, and task offloading ratio of the vehicle formation; The second Markov decision process is solved to obtain parameters corresponding to the vehicle spacing of the vehicle formation.

7. The resource management method of vehicle formation according to claim 6, characterized in that: The state of the first Markov decision process is defined as key information reflecting the communication environment and task offloading, the action is defined as the decision corresponding to the transmit power, beam width and task offloading ratio, and the reward function includes the function corresponding to the task processing delay and the task delay constraint; The state of the second Markov decision process is defined as the physical state of the vehicle and the overall condition of the vehicle formation, the action is defined as the decision corresponding to the vehicle spacing, and the reward function includes minimization of the average vehicle spacing, minimization of the task offloading delay, and functions corresponding to the safety constraints.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the resource management method for a combined vehicle formation as claimed in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the resource management method of a combined vehicle formation as claimed in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the resource management method of a combined vehicle formation as claimed in any one of claims 1 to 7 is implemented.

Images