V2X scene-oriented general inductance algorithm fusion heterogeneous resource allocation method

Abstract

The invention belongs to the field of wireless communication in the Internet of Vehicles, and particularly relates to a V2X scene-oriented communication calculation fusion heterogeneous resource allocation method. The method comprises the following steps: firstly, constructing an Internet of Vehicles communication, sensing and calculation integrated system which comprises a vehicle, a base station, a road side unit and an edge calculation node; and secondly, comprehensively considering path loss, a multipath effect and a Doppler effect, and performing channel modeling on a 5G-Uu link, a C-V2X PC5 link and a Wi-Fi link by respectively adopting a macro cellular model, an Internet of Vehicles direct connection channel model and an IEEE 802.11 p model. Then, constructing a reinforcement learning resource optimization framework based on time scale layering, wherein an upper layer realizes proportional dynamic allocation of spectrum resources through dynamic resource pool division; and the lower layer jointly optimizes perception, communication and computing resource allocation in a set resource area. According to the method, on the premise of meeting various service quality constraints, the sum of the calculation cost, the communication cost and the penalty cost can be minimized, the total cost of the communication, sensing and calculation integrated architecture is minimized, and the resource utilization rate is improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-everything (V2X) communication technology, and in particular to a heterogeneous resource allocation method that integrates sensing, computing, and communication for V2X scenarios, which is especially suitable for real-time task assurance in V2X scenarios. Background Technology

[0002] With the continuous evolution of sixth-generation mobile communication technology, the Internet of Vehicles (IoV) is accelerating its development towards the integration of sensing, communication, and computing. Future intelligent IoVs not only require highly reliable, low-latency communication, but also need to provide high-precision environmental perception and efficient data processing capabilities to support diverse applications such as autonomous driving and collaborative decision-making. To this end, vehicles typically integrate multiple types of sensors, advanced communication modules, and local / edge computing units to achieve real-time environmental perception, information interaction, and intelligent processing.

[0003] In traditional vehicle-to-everything (V2X) systems, communication, sensing, and computing resources often employ independent scheduling and allocation mechanisms, making it difficult to adapt to the dynamic resource demands of multi-service concurrent scenarios. Particularly regarding spectrum resources, existing solutions often use static allocation based on single metrics such as communication throughput or sensing accuracy, lacking overall optimization for spectrum efficiency. For example, in high-precision sensing tasks, a large amount of spectrum resources may be occupied by sensing functions, leading to a decline in communication quality; while in communication-intensive scenarios, spectrum resources fail to be dynamically adjusted according to the real-time demands of sensing and computing tasks, resulting in uneven resource utilization and impacting overall system performance.

[0004] Furthermore, most existing vehicle-to-everything (V2X) resource scheduling methods fail to fully consider the collaborative allocation and dynamic sharing mechanisms of spectrum resources in the context of integrated communication, sensing, and computing. Spectrum, as a key resource shared by communication and sensing, directly impacts end-to-end system performance through its allocation strategy. Traditional communication-centric spectrum allocation models are ill-suited to the high bandwidth and low latency requirements of sensing tasks, and fail to effectively combine computing resource processing status for joint optimization. This results in low spectrum utilization, increased inter-service interference, and an inability to meet the demands of 6G V2X for efficient integration of multi-dimensional resources.

[0005] With the advancement of the integrated communication, sensing, and computing architecture, the efficient reuse and dynamic allocation of spectrum resources have become crucial for improving system performance. By constructing a spectrum resource scheduling algorithm oriented towards the fusion of communication, sensing, and computing, it is possible to improve sensing accuracy and computational efficiency while ensuring the quality of communication services, thereby optimizing the overall system resource performance. Therefore, there is an urgent need for a collaborative optimization method centered on spectrum resource allocation, capable of dynamically adjusting spectrum usage strategies based on real-time task requirements and channel state information, to support the reliable operation of 6G vehicle-to-everything (V2X) networks in multi-service, highly dynamic environments.

[0006] This study aims to address the aforementioned needs by proposing a spectrum resource optimization allocation algorithm for integrated sensor-computer communication vehicles, thereby achieving efficient utilization of spectrum resources and synergistic improvement of multi-dimensional service quality, and providing key technical support for the development of intelligent vehicle networks. Summary of the Invention

[0007] To address the aforementioned issues, this invention provides a method for allocating heterogeneous resources through sensory computing fusion in V2X scenarios, comprising the following specific steps:

[0008] S1. Construct a vehicle-to-everything (V2X) integrated sensing and computing system, including: vehicles (V), base stations (BS), roadside units (RSU), and mobile edge computing nodes (MEC). Vehicles and RSUs communicate through vehicle-to-infrastructure (V2I) links, and vehicles communicate with each other through vehicle-to-vehicle (V2V) links. The data flow process in the system is as follows: vehicles send sensing data to other vehicles through V2V links, and simultaneously send sensing data to RSUs through V2I links; RSUs process the received data and feed the processing results back to the vehicles.

[0009] S2. Construct heterogeneous network channel models. For 5G-Uu links, C-V2X PC5 links and Wi-Fi links, respectively, adopt the corresponding channel models to calculate channel gain. The channel models comprehensively consider path loss, multipath effect and Doppler effect.

[0010] S3. Optimize the integrated sensing and computing network by establishing a time-scale layered reinforcement learning optimization framework. The upper layer dynamically allocates spectrum resources to each network proportionally through dynamic resource pool partitioning decisions. The lower layer performs joint allocation of sensing, communication, and computing resources within the partitioned resource areas. Under various quality of service constraints, the minimum cost of the integrated communication, sensing, and computing architecture is obtained by minimizing the accumulation of computing cost, communication cost, and penalty cost.

[0011] The beneficial effects of this invention are:

[0012] This invention realizes a hierarchical reinforcement learning optimization framework for heterogeneous resource allocation in the Internet of Vehicles (IoV) based on dual-timescale fusion of sensing and computing. Based on multi-agent reinforcement learning theory, it proposes a time-scale decoupled hierarchical decision-making architecture. Through upper-layer dynamic resource pool management, it achieves flexible allocation of heterogeneous network resources such as 5G-Uu and C-V2X. At the lower layer, it jointly optimizes communication, sensing, and computing resources to ensure system service quality and resource utilization efficiency. Hierarchical distributed decision-making reduces optimization complexity. Combining offline training with online inference improves real-time decision-making performance. Furthermore, utilizing the cost model designed in this invention, it significantly reduces the total system operating cost and improves resource utilization in multi-task concurrent scenarios. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a system model for a heterogeneous resource allocation method for V2X scenarios using synergistic computing in an embodiment of the present invention.

[0014] Figure 2 This is a flowchart of a heterogeneous resource allocation method for synergistic computing fusion in V2X scenarios, as described in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Based on the above requirements, the present invention provides an embodiment of a heterogeneous resource allocation method for the integration of sensing and computing in the context of limited spectrum resources for vehicle-to-everything (V2X) systems. The V2X system constructed in this method involves four types of entities: vehicles, base stations, roadside units, and edge computing nodes.

[0017] In the embodiments provided by this invention, considering the resource allocation and cost issues under the integrated sensing and computing architecture of the vehicle-to-everything (V2X) network, a system architecture for heterogeneous resource allocation in the context of sensing and computing fusion in V2X scenarios is constructed, such as... Figure 1 As shown, where:

[0018] As a perception and communication terminal, the vehicle is equipped with multiple types of sensors and a V2X communication module, which are responsible for collecting environmental perception data and generating computing tasks, while also having a certain local computing capability.

[0019] Base stations form the core of cellular network coverage, integrate MEC functions, are responsible for macro-level spectrum resource management and cross-RSU resource coordination, and provide high-speed backhaul links and cloud computing access services for vehicles and RSUs.

[0020] Roadside units are deployed at key road nodes and have communication, sensing, and edge computing capabilities. They are responsible for receiving and processing data sent by vehicles, providing regional sensing fusion and computing services, and working with base stations to complete resource allocation.

[0021] Vehicles and RSUs communicate via V2I links, while vehicles communicate with each other via V2V links. The data flow in the system is as follows: vehicles send sensing data to other vehicles via V2V links and simultaneously send sensing data to the RSUs via V2I links. The RSUs process the received data and feed the results back to the vehicles. Additionally, the RSUs can send data to MEC nodes for further processing. Task load varies over time, including the generation of new tasks, task completion, and task priority adjustments. This variation in task load can be described using a task queue model, where the number of tasks and the complexity of each task queue change over time.

[0022] like Figure 2 As shown, a method for allocating heterogeneous resources through sensory computing fusion in V2X scenarios includes the following steps:

[0023] S1. Construct a vehicle-to-everything (V2X) communication, sensing, and computing integrated system, including: vehicles, base stations, roadside units (Roadside Units), and edge computing nodes. Vehicles and Roadside Units (RSUs) communicate via V2I links, and vehicles communicate with each other via V2V links. The data flow process in the system is as follows: vehicles send sensing data to other vehicles via V2V links, and simultaneously send sensing data to RSUs via V2I links; RSUs process the received data and feed back the processing results to the vehicles.

[0024] S2. Construct heterogeneous network channel models. For 5G-Uu links, C-V2X PC5 links and Wi-Fi links, respectively, adopt the corresponding channel models to calculate channel gain. The channel models comprehensively consider path loss, multipath effect and Doppler effect.

[0025] S3. Optimize the integrated sensing and computing network by establishing a time-scale layered reinforcement learning optimization framework. The upper layer dynamically allocates spectrum resources to each network proportionally through dynamic resource pool partitioning decisions. The lower layer performs joint allocation of sensing, communication, and computing resources within the partitioned resource areas. Under various quality of service constraints, the minimum cost of the integrated communication, sensing, and computing architecture is obtained by minimizing the accumulation of computing cost, communication cost, and penalty cost.

[0026] Preferably, each synesthetic computing node in S1 involves the design of its communication process and the construction of its task queue model. This aims to achieve this through reasonable communication processes and task queue management. Specifically, it includes:

[0027] S11. The communication process of each sensing node involves the integration of communication gateways for multiple wireless access technologies, supporting 5G, C-V2X PC5, and Wi-Fi. The network access decision variable is a. t,k ∈{0,1}, k is the ID of the network resource, t is the ID of the task in the task queue, and a t,k When a = 1, it means that task t is allocated to the resource pool with network resource k, and a t,k When α = 0, the opposite is true, where the decision variable for the resource pool allocation ratio is α. k (s)∈[0,1], representing the proportion of resources allocated to network technology k in time slot s, where the parameters of the resource pool are defined, n is the number of the communication resources divided into subcarriers in the frequency domain, and a resource block (n,s) can be determined by using n and s as a two-dimensional grid;

[0028] S12. The task queue model consists of two types: a computation task queue and a communication data queue. Each item in the computation task queue represents a task to be processed, and its size is assumed to be C, which is the computational cost C required by the task t. t The decision is made in which each item in the communication data queue represents a data packet to be transmitted, assuming its size is determined by the amount of data D in task t. t,in Or D t,out Decision, of which D t,in D represents the amount of original data that needs to be transmitted to the computing node for processing in the perception task t. t,out This refers to the amount of result data that needs to be returned to the requesting vehicle after being transmitted to the computing node.

[0029] Preferably, the specific formula for calculating the channel gain in S2 is as follows: Where H u→v,n,s L is the complex channel gain from the transmitting node u to the receiving node v in resource block (n,s), and its parameter value depends on the network resource settings used. u→v The path loss is from the transmitting node u to the receiving node v. PL0 represents the path loss at a reference distance d0, α is the path loss exponent depending on the environment (primarily considering urban environments), d is the distance between the transmitting node u and the receiving node v, and α is the path loss exponent depending on the environment (primarily considering urban environments). α is taken as a Monte Carlo random number within (3,4). F shadow To quantify the slow signal variation caused by obstacles during the transmission from transmitter u to receiver v, a log-normal distribution is used here. For F shadow standard deviation This refers to the small-scale fading between the transmitting node u and the receiving node v.

[0030] Preferably, the channel model in S2 is divided into a transmit signal model and a receive signal model, wherein X t,n,s Let ∈{0,1} be the signal transmitted by task t on subcarrier n and time slot s, where the expression for the signal-to-interference-plus-noise ratio (SINR) of the signal received by node v from node u is: Where S t,n,s The signal power sent by task t on resource block (n,s) is expressed as follows: Let I be the transmit power allocated in resource block (n,s) for transmitting task t, and N be the noise power in resource block (n,s), where I t,n,s It is the interference power experienced by task t on resource block (n,s), expressed as: Where H u′→v,n,s For the channel gain from other interfering nodes u′ to the receiving node v. This is the transmit power used on resource block (n,s) to interfere with node u′ sending tasks.

[0031] Preferably, the continuous decision variable α for the resource pool allocation ratio in S3 is... k (s)∈[0,1], this decision variable acts on the upper-level slow decision of hierarchical optimization. Every time interval τ, based on the predicted network load, the optimal resource pool allocation ratio for a future period is optimized. The observed state S(τ) is set as the average queue length, historical load, and channel condition trend of each network, and is expressed as S(τ)=[Q(τ-1),Λ(τ-1),H(τ-1),ΔH(τ-1)]. T Where Q(τ-1) is the average queue length of each network in the previous cycle, expressed as Within one decision period τ, the average total queue length of network k is expressed as: k∈{5G,CV2X,WiFi}; where S τ The number of fast time slots contained in the period τ. The sum of the total queue data for all tasks using network k in fast time slot s, where The historical load of each network is recorded, and H(τ-1) is the average channel gain of each network in the previous period. Within a decision period τ, the average channel gain of network k is expressed as... N k H represents the number of subcarriers allocated to network k during period τ. k,n,s Let k be the channel gain of network k in resource block (n,s), and Region k Let denot be the resource pool corresponding to network technology k, and ΔH(τ-1) be the channel gain change trend of each network in the previous period, obtained by difference, where action α(τ)=[α 5G (τ),α CV2X (τ),αWiFi [τ] represents the resource pool allocation ratio, and the reward is set based on the time interval T of feedback from the lower-level agent. slow The negative weighted sum of costs is expressed as R(τ) = -C total (τ+1), the optimization objective is to maximize the expected cumulative reward, which is averaged over all possible time slots according to the action, and is expressed as: Where γ∈[0,1] is the discount factor.

[0032] Preferably, the cost expression in S3 is C total =C comp +C comu +C pena C comp The computational cost of all tasks across all nodes is expressed as C. comp =∑t∑m[x t,m ·(E m ·T t,comp )+(P m ·C t )], where x t,m For binary unloading of decision variables, whether decision task t is unloaded to compute node m for execution, m∈{V,RSU,MEC,BS}, where E m T represents the energy cost per unit time for running on node m. t,comp Let be the computation time of task t on node m, denoted as f t,m The computing resources allocated to the current node m, where P m Let C be the unit computational cost of node m, i.e., the direct cost of renting computing power. comu The cost incurred for data transfer using V2I or V2V is denoted as C. comu =∑t[((T) t,up +T t,down )·C time +(D t,in +D t,out )·C data )], where T t,up To make D t,in The time taken to transmit to the computing node is expressed as Where R t Let be the average effective transmission rate obtained by task t during the entire communication process, denoted as Where ρ t,n,s For binary communication routing decision variables, the decision is whether to allocate the resource block with subcarrier n and time slot s to task t, r. t,n,s Let r be the instantaneous transfer rate achieved by task t on resource block (n,s). t,n,s =Bsub ·log2(1+SINR t,n,s ), where B sub For a single subcarrier, the bandwidth constant is , where For download time, R t ' is the download speed of task t, expressed as Where C time The cost is calculated per unit time, based on duration, starting when the task begins transmission, where C data The unit data cost, where C pena The penalty cost is expressed as C. pena =∑t[max(0,(T)] t,total -T t,max ))·R pena ], where T t,total Let T be the total completion time of task t. t,total =T t,comm +T t,comp T t,comm Let R be the communication time for task t, including upload and download times, where R is the communication time for task t. pena The penalty amount is calculated per unit of time.

[0033] Preferably, the lower-level optimization model in S3 is a model using the MADDPG algorithm, where the state is α determined by the upper-level decision. k (s), D t,in D t,out C t H u→v,n,s The action is x t,m ρ t,n,s , f t,m The reward is C. total .

[0034] Preferably, the constraints in S3 are C1 to C10, where constraint C1 represents the constraint resource ratio. Constraint C2 indicates the uniqueness of the constraint unloading decision. Constraint C3 represents a constraint on computational resource capacity. The maximum computing resources that can be allocated to the current node m are defined by constraint C4, which indicates a constraint on the coupling of computing power allocation. Constraint C5 indicates that the transmit power is constrained to be non-negative. Let resource block (n,s) be the maximum transmit power that can be allocated to the data of transmission task t. Constraint C6 represents an orthogonality constraint. Constraint C7 indicates that the task can only use resource blocks from the resource pool corresponding to its chosen network technology k. Constraint C8 indicates that if task t is computed locally, then the task does not need to use any network for transmission. Constraint C9 represents the end-to-end latency constraint.

[0035] This invention addresses the problem of uneven resource utilization in vehicle-to-everything (V2X) networks by designing a heterogeneous resource allocation optimization method that integrates sensing, perception, and computing resources in environments with limited spectrum resources. The proposed method leverages the differences in communication and computing resources among heterogeneous nodes such as vehicles, RSUs, MECs, and base stations, achieving efficient resource utilization through a three-pronged mechanism: First, it constructs a cross-domain joint resource allocation model to comprehensively optimize spectrum, sensing, and computing resources, significantly reducing the complexity of multi-dimensional resource management. Second, it designs a dynamic decision-making framework based on dual time scales, with the upper layer performing macro-level spectrum partitioning of the heterogeneous network and the lower layer implementing fine-grained resource block and computing power scheduling, thus balancing decision-making efficiency and accuracy. Third, it introduces a lightweight online inference mechanism, generating optimization schemes in real time through a pre-trained policy network, significantly reducing online computational overhead. Simulation results demonstrate that this invention effectively improves resource utilization efficiency in heterogeneous environments and significantly reduces the total system cost while ensuring various quality of service constraints.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for allocating heterogeneous resources through sensory computing fusion in V2X scenarios, characterized in that, Includes the following steps: S1. Construct a vehicle-to-everything (V2X) integrated sensing and computing system, including: vehicles (V), base stations (BS), roadside units (RSU), and mobile edge computing nodes (MEC). Vehicles and RSUs communicate through vehicle-to-infrastructure (V2I) links, and vehicles communicate with each other through vehicle-to-vehicle (V2V) links. The data flow process in the system is as follows: vehicles send sensing data to other vehicles through V2V links, and simultaneously send sensing data to RSUs through V2I links; RSUs process the received data and feed the processing results back to the vehicles. S2. Construct heterogeneous network channel models. For 5G-Uu links, C-V2X PC5 links and Wi-Fi links, respectively, adopt the corresponding channel models to calculate channel gain. The channel models comprehensively consider path loss, multipath effect and Doppler effect. S3. Optimize the integrated sensing and computing network by establishing a time-scale layered reinforcement learning optimization framework. The upper layer dynamically allocates spectrum resources to each network proportionally through dynamic resource pool partitioning decisions. The lower layer performs joint allocation of sensing, communication, and computing resources within the partitioned resource areas. Under various quality of service constraints, the minimum cost of the integrated communication, sensing, and computing architecture is obtained by minimizing the accumulation of computing cost, communication cost, and penalty cost.

2. The method for heterogeneous resource allocation through sensory computing fusion for V2X scenarios according to claim 1, characterized in that, The S1 section involves the design of communication processes and the construction of task queue models for each synesthetic computing node. The aim is to achieve this through reasonable communication processes and task queue management. Specifically, this includes: S11. The communication process of each sensing node involves the integration of communication gateways for multiple wireless access technologies, supporting 5G, C-V2X PC5, and Wi-Fi. The network access decision variable is a. t,k ∈{0,1}, k is the ID of the network resource, t is the ID of the task in the task queue, and a t,k When a = 1, it means that task t is allocated to the resource pool with network resource k, and a t,k When α = 0, the opposite is true, where the decision variable for the resource pool allocation ratio is α. k (s)∈[0,1], representing the proportion of resources allocated to network technology k in time slot s, where the parameters of the resource pool are defined, n is the number of the communication resources divided into subcarriers in the frequency domain, and a resource block (n,s) can be determined by using n and s as a two-dimensional grid; S12. The task queue model consists of two types: computation task queues and communication data queues. Each item in the computation task queue represents a task to be processed, and its size is assumed to be C, which is the computational cost C required for that task t. t The decision is made in which each item in the communication data queue represents a data packet to be transmitted, assuming its size is determined by the amount of data D in task t. t,in Or D t,out Decision, of which D t,in D represents the amount of original data that needs to be transmitted to the computing node for processing in the perception task t. t,out This refers to the amount of result data that needs to be returned to the requesting vehicle after being transmitted to the computing node.

3. The method for heterogeneous resource allocation through sensory computing fusion for V2X scenarios according to claim 1, characterized in that, The specific formula for calculating the channel gain in S2 is as follows: Where H u→v,n,s L is the complex channel gain from the transmitting node u to the receiving node v in resource block (n,s), and its parameter value depends on the network resource settings used. u→v The path loss is from the transmitting node u to the receiving node v. PL0 represents the path loss at a reference distance d0, where d is the distance between the transmitting node u and the receiving node v, and α is the path loss exponent, which depends on the environment. Here, we mainly consider the urban environment, and α is taken as a Monte Carlo random number within (3,4). F shadow To quantify the slow signal variation caused by obstacles during the transmission from transmitter u to receiver v, a log-normal distribution is used here. For F shadow standard deviation This refers to the small-scale fading between the transmitting node u and the receiving node v.

4. The method for heterogeneous resource allocation through sensory computing fusion for V2X scenarios according to claim 1, characterized in that, The channel model in S2 is divided into a transmit signal model and a receive signal model, wherein X t,n,s Let ∈{0,1} be the signal transmitted by task t on subcarrier n and time slot s, where the expression for the signal-to-interference-plus-noise ratio (SINR) of the signal received by node v from node u is: Where S t,n,s The signal power sent by task t on resource block (n,s) is expressed as follows: Let I be the transmit power allocated in resource block (n,s) for transmitting task t, and N be the noise power in resource block (n,s), where I t,n,s It is the interference power experienced by task t on resource block (n,s), expressed as: Where H u′→v,n,s For the channel gain from other interfering nodes u′ to the receiving node v. It is the transmit power used on resource block (n,s) to interfere with node u′ sending tasks.

5. A method for allocating heterogeneous resources through sensory computing fusion in V2X scenarios according to claim 1, characterized in that, The continuous decision variable α for the resource pool allocation ratio in S3 k (s)∈[0,1], this decision variable acts on the upper-level slow decision of hierarchical optimization. Every time interval τ, based on the predicted network load, the optimal resource pool allocation ratio for a future period is optimized. The observed state S(τ) is set as the average queue length, historical load, and channel condition trend of each network, and is expressed as S(τ)=[Q(τ-1),Λ(τ-1),H(τ-1),ΔH(τ-1)]. T Where Q(τ-1) is the average queue length of each network in the previous cycle, expressed as Within one decision period τ, the average total queue length of network k is expressed as: k∈{5G,CV2X,WiFi}; where S τ The number of fast time slots contained in the period τ. The sum of the total queue data for all tasks using network k in fast time slot s, where The historical load of each network is recorded, and H(τ-1) is the average channel gain of each network in the previous period. Within a decision period τ, the average channel gain of network k is expressed as... N k H represents the number of subcarriers allocated to network k during period τ. k,n,s Let k be the channel gain of network k in resource block (n,s), and Region k Let denot be the resource pool corresponding to network technology k, and ΔH(τ-1) be the channel gain change trend of each network in the previous period, obtained by difference, where action α(τ)=[α 5G (τ),α CV2X (τ),α WiFi [τ] represents the resource pool allocation ratio, and the reward is set based on the time interval T of feedback from the lower-level agent. slow The negative weighted sum of costs is expressed as R(τ) = -C total (τ+1), the optimization objective is to maximize the expected cumulative reward, which is averaged over all possible time slots according to the action, and is expressed as: Where γ∈[0,1] is the discount factor.

6. The method for heterogeneous resource allocation through sensory computing fusion for V2X scenarios according to claim 1, characterized in that, The cost expression in S3 is C. total =C comp +C comu +C pena C comp The computational cost of all tasks across all nodes is expressed as C. comp =∑t∑m[x t,m ·(E m ·T t,comp )+(P m ·C t )], where x t,m For binary unloading of decision variables, whether decision task t is unloaded to compute node m for execution, m∈{V,RSU,MEC,BS}, where E m T represents the energy cost per unit time for running on node m. t,comp Let be the computation time of task t on node m, denoted as f t,m The computing resources allocated to the current node m, where P m Let C be the unit computational cost of node m, i.e., the direct cost of renting computing power. comu The cost incurred for data transfer using V2I or V2V is denoted as C. comu =∑t[((T) t,up +T t,down )·C time +(D t,in +D t,out )·C data )], where T t,up To make D t,in The time taken to transmit to the computing node is expressed as Where R t Let be the average effective transmission rate obtained by task t during the entire communication process, denoted as Where ρ t,n,s For binary communication routing decision variables, the decision is whether to allocate the resource block with subcarrier n and time slot s to task t, r. t,n,s Let r be the instantaneous transfer rate achieved by task t on resource block (n,s). t,n,s =B sub ·log2(1+SINR t,n,s ), where B sub For a single subcarrier, the bandwidth constant is , where For download time, R t ' is the download speed of task t, expressed as Where C time The cost is calculated per unit time, based on duration, starting when the task begins transmission, where C data The unit data cost, where C pena The penalty cost is expressed as C. pena =∑t[max(0,(T)] t,total -T t,max ))·R pena ], where T t,total Let T be the total completion time of task t. t,total =T t,comm +T t,comp T t,comm Let R be the communication time for task t, including upload and download times, where R is the communication time for task t. pena The penalty amount is calculated per unit of time.

7. The method for optimizing heterogeneous resource allocation through sensory computing fusion in V2X scenarios according to claim 1, characterized in that, The lower-level optimization model in S3 is a model using the MADDPG algorithm, where the state is α determined by the upper-level decision. k (s), D t,in D t,out C t H u→v,n,s The action is x t,m ρ t,n,s , f t,m The reward is C. total .

8. A method for allocating heterogeneous resources through sensory computing fusion in V2X scenarios according to claim 1, characterized in that, The constraints in S3 are C1 to C10, where constraint C1 represents the constraint resource ratio. Constraint C2 indicates the uniqueness of the constraint unloading decision. Constraint C3 represents a constraint on computational resource capacity. The maximum computing resources that can be allocated to the current node m are defined by constraint C4, which indicates a constraint on the coupling of computing power allocation. Constraint C5 indicates that the transmit power is constrained to be non-negative. Let resource block (n,s) be the maximum transmit power that can be allocated to the data of transmission task t. Constraint C6 represents an orthogonality constraint. Constraint C7 indicates that the task can only use resource blocks from the resource pool corresponding to its chosen network technology k. Constraint C8 indicates that if task t is computed locally, then the task does not need to use any network for transmission. Constraint C9 represents the end-to-end latency constraint.

Images