Calculation unloading system and method oriented to car networking service

Abstract

The invention discloses a computing unloading system and method oriented to car networking service. The method comprises the following steps: carrying out demand analysis on an Internet of Vehicles service to obtain a time-sensitive demand type and a computing power demand type of the Internet of Vehicles service; decomposing the service to obtain a plurality of sub-tasks, and determining an execution sequence, a computing power demand and a computing resource of each sub-task according to the time-sensitive demand type and the computing power demand type; determining a distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type; and according to the computing power demand type, the execution time determined based on the subtask execution sequence, and the execution energy consumption determined based on the computing power demand of the subtasks and the computing resources, taking the service delay and the execution energy consumption as optimization targets, and obtaining an optimal computing unloading scheme under the distributed computing power fusion architecture based on a grey wolf algorithm of a simulated annealing mechanism. Layered deployment is performed on the computing power resources based on the car networking service requirements, so that real-time collaboration of the computing power resources and efficient response of services are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle networking, and in particular, to a computing offloading system and method for vehicle networking services. Background Art

[0002] An intelligent connected vehicle refers to the organic combination of vehicle networking and intelligent vehicles. It is equipped with advanced on-vehicle sensors, controllers, actuators and other devices, and integrates modern communication and network technologies to realize the exchange and sharing of intelligent information among vehicles, people, roads, backends, etc., to achieve safe, comfortable, energy-saving and efficient driving, and ultimately can replace humans to operate a new generation of vehicles.

[0003] In the wave of the development of intelligent connected vehicles, edge computing (EC), a completely distributed processing architecture, supports sinking the computing power of cloud data centers to the edge and terminal to provide real-time high-quality edge intelligent services nearby, improve network service performance, and can meet the key requirements of intelligent terminals in aspects such as agile connection, real-time services, and data optimization. Vehicle networking + edge computing can effectively utilize the "field-level" intelligent services provided by edge computing to improve the practical problems faced by vehicle networking applications, such as the real-time and reliability problems of terminal applications, and the sharing problems of intelligent information data such as vehicles, roads, and people. It provides a connection channel for a large number of complex heterogeneous devices and networks in the vehicle networking field, and is an effective means to promote the development of the intelligent connected vehicle industry.

[0004] However, at the present stage in the process of the transformation of vehicle networking + edge computing, in the process of effectively coordinating intelligent terminals (such as mobile devices, in-vehicle CPUs, intelligent traffic lights, cameras, etc.) and edge nodes (such as roadside units) to provide intelligent services, the following problems and challenges are faced: there are a wide variety of heterogeneous terminal devices and edge node resources, and it is difficult to effectively coordinate across different management domains. There are many and miscellaneous on-site devices, and there is no unified paradigm for the complete interconnection and intercommunication of network communication protocols. The application scenarios of vehicle networking are variable, the on-site edge devices are heterogeneous, and there is a lack of an effective heterogeneous resource management mechanism. At the same time, vehicle networking services, such as intelligent obstacle avoidance, intelligent early warning and other applications, have high requirements for high real-time performance. How to perform distributed collaborative scheduling of resources in the environment and effectively complete computing offloading is a huge challenge faced by the vehicle networking system.

[0005] Currently, many studies have proposed deep learning-based methods to solve the computing offloading strategy for 5G vehicle-to-everything (V2X) edge services. Such methods have high resource requirements and high costs themselves. Moreover, when there are many terminals and edge server nodes, the state space and action space of the problem will both increase exponentially, resulting in the curse of dimensionality, which leads to slow training speed and cannot meet the high real-time requirements of services. Therefore, how to develop an efficient computing offloading method and a distributed computing power fusion management mechanism that can balance time and cost is one of the urgent problems to be solved in current V2X services. Summary of the Invention

[0006] Aiming at the problems in existing V2X systems, such as the difficulty in effectively coordinating distributed computing power and the inability of the computing offloading strategy to meet high real-time requirements, the present invention proposes a computing offloading system and method for V2X services.

[0007] A computing offloading system for V2X services to achieve one of the objectives of the present invention includes:

[0008] A service demand analysis module: used to perform multi-dimensional demand analysis on V2X services to obtain the time-sensitive demand type and computing power demand type of each V2X service;

[0009] A service process construction module: used to decompose V2X services into multiple independently executable subtasks according to the execution process of each V2X service; and determine the execution sequence, computing power demand, and computing resources of each subtask according to the time-sensitive demand type and computing power demand type of the V2X service.

[0010] A computing power fusion module: used to determine a distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type of V2X services; each level represents a different computing power demand type.

[0011] A computing offloading module: taking service latency and execution energy consumption as optimization objectives, using a grey wolf algorithm based on the simulated annealing mechanism to calculate the optimal computing offloading scheme under the distributed computing power fusion architecture according to the computing power demand type of each V2X service, the execution time of the V2X service determined based on the execution sequence of subtasks, and the execution energy consumption of the V2X service determined based on the computing power demand and computing resources of subtasks; the optimal computing offloading scheme uses the distributed computing power fusion architecture to achieve real-time scheduling of multi-level computing power resources.

[0012] A further technical solution includes: in the service demand analysis module, the method for obtaining the time-sensitive demand type and computing power demand type of each V2X service includes:

[0013] Analyze each vehicle networking service according to the data flow, computing process, and network interaction dimensions, and extract key requirements; the key requirements include: data-related requirements, computing-related requirements, and network-related requirements;

[0014] According to the network-related requirements, determine the time-sensitive requirement types for each vehicle networking service; for example, vehicle networking services with extremely high requirements for response time, where even a slight delay may lead to serious consequences, are classified as extremely time-sensitive; vehicle networking services with certain requirements for response time but allowing relatively short delays are classified as moderately time-sensitive; vehicle networking services with lower requirements for response time are classified as not time-sensitive;

[0015] According to the data-related requirements and computing-related requirements, determine the computing power requirement types for each vehicle networking service; the computing power requirement types include data-intensive, computing-intensive, and network-intensive.

[0016] The technical effects of the above technical solutions include: avoiding the one-sidedness of single-dimensional analysis through data flow, computing process, and network interaction analysis; classifying service types based on network-related requirements, providing time constraints for task offloading strategies; determining the computing power type by combining data volume and computing complexity, ensuring the matching of tasks and computing power resources, and improving computing efficiency.

[0017] Further technical solutions include: the time-sensitive requirement types of vehicle networking services include: extremely time-sensitive, moderately time-sensitive, and not time-sensitive; the ranking of response time requirements from high to low is: extremely time-sensitive service > moderately time-sensitive > not time-sensitive; the method for decomposing a vehicle networking service into multiple independently executable subtasks includes:

[0018] For extremely time-sensitive (H) services: decompose the service into a strictly serial subtask chain;

[0019] For moderately time-sensitive (M) services: decompose into multiple subtask sequences allowing short-delay elastic scheduling, and subtasks can be dynamically merged and delayed for execution;

[0020] For not time-sensitive (L) services: decompose into multiple subtasks for asynchronous batch processing, and subtasks support task interruption and idle resource scheduling.

[0021] Further technical solutions include: each vehicle networking service is represented by a directed acyclic graph (V i , E_edge i ), where V i represents the subtask set of the i-th vehicle networking service s i , V i = {v i,1 , vi,2 , …, v i,n-1 , v i,n}, E_edge i = {e i |e i = (v i,m , v i,n )} represents the set of edges, where the i-th edge points from vertex v i,m to v i,n , indicating that subtask v i,n must be executed after subtask v i,m .

[0022] The technical effects of the above technical solution include: The solution of splitting the service into independent subtasks supports distributed parallel processing, shortens the overall execution time, provides fine-grained control for dynamic scheduling, and adapts to the dynamic changes of services in the vehicle network; The execution order of subtasks is clarified through a directed acyclic graph, avoiding task conflicts and ensuring the logical correctness of the service process.

[0023] Further technical solutions include: The method for determining the execution time and execution energy consumption of each vehicle network service includes:

[0024] When the computing power demand type of the vehicle network service is provided by the first computing power node network composed of terminal devices, the execution time and execution energy consumption of each subtask of the vehicle network service are respectively: T i,j = c_q i,j / c i , E i,j = c_q i,j × pc i / c i ;

[0025] When the computing power demand type of the vehicle network service is provided by the second computing power node network composed of edge devices, the execution time and execution energy consumption of each subtask of the vehicle network service are respectively: T i,j = c_q i,j / c i + c_q i,j / n i ;

[0026] When the computing power demand type of the vehicle network service is provided by the cloud computing power node network, the execution time and execution energy consumption of each subtask of the vehicle network service are respectively: T i,j = c_q i,j / c i + c_q i,j / n i , E i,j = c_q i,j × pci / c i +c_q i,j ×pt i / n i ;

[0027] T i,j and E i,j respectively represent the execution time and execution energy consumption of the j-th subtask of the i-th vehicle networking service; c_q i,j represents the amount of data to be calculated for the j-th subtask of the i-th vehicle networking service; c i represents the computing power of the computing power type to which the i-th vehicle networking service belongs; n i represents the network transmission capacity of the computing power type to which the i-th vehicle networking service belongs; pc i represents the computing power of the computing power type to which the i-th vehicle networking service belongs; pt i represents the transmission power of the computing power type to which the i-th vehicle networking service belongs;

[0028] Add up the execution times of all subtasks to obtain the execution time of this vehicle networking service;

[0029] Add up the execution energy consumptions of all subtasks to obtain the execution energy consumption of this vehicle networking service.

[0030] The technical effects of the above technical solutions include: establishing execution time and energy consumption models for terminal, edge, and cloud computing power nodes respectively, and considering parameters such as computing power, network transmission capacity, and power, ensuring the close combination of the model with the actual hardware performance and improving the decision-making reliability; at the same time, through the accumulation of subtask execution time and energy consumption, the global performance evaluation of the vehicle networking service is realized.

[0031] Further technical solutions include: The calculation method of the optimal computing offloading scheme includes:

[0032] Set the position of each gray wolf to (s i , r_type j ), where s i represents the vehicle networking service, and r_type j represents the computing power type assigned to the vehicle networking service;

[0033] Set the objective function value of the optimal gray wolf to f(s i , r_type j ) = ∑T i / ∑t_d i +∑E i / ∑e_d i ; where, ∑t_d i represents the maximum allowable execution time of all current vehicle networking tasks; ∑e_d iIndicates the system energy consumption of all current Internet of Vehicles tasks under the maximum delay condition;

[0034] Initialize all parameters of the simulated annealing algorithm, including population size, maximum number of iterations, initial temperature value, and randomly initialize the positions of individuals in the gray wolf population, calculate the objective function value of the gray wolf population, and record the optimal position;

[0035] Update the gray wolf's position according to the simulated annealing mechanism, and define the probability of accepting the new position as The position is selected, where f and f' are the objective function values ​​of the current position and the new position respectively, and T(h) represents the temperature parameter in the simulated annealing process; if p is greater than the set value, the gray wolf position is updated to the new position;

[0036] Determine whether the iteration condition is met, if not, repeat the above steps of updating the gray wolf position; if met, output the optimal solution, which is the optimal computation offloading solution.

[0037] The technical effects of the above technical scheme include: the swarm intelligence of the Grey Wolf algorithm and the global search capability of simulated annealing can effectively balance the conflict between delay and energy consumption, avoiding local optimality; gradually narrowing the search range through the annealing mechanism, improving the convergence speed and accuracy of the algorithm; based on the Grey Wolf position coding (service-computing power type mapping), rapid traversal of the solution space is achieved, which can well respond to the high real-time requirements of the Internet of Vehicles.

[0038] A further technical solution includes: the Internet of Vehicles service i Includes: i =(t i ,q i ,t_type i ,c_type i ,t_d i ,c_q i );t i represents the request service time of the i-th service, q i Indicates the service initiator device requesting the i-th service; t_type i Indicates the time-sensitive type of the i-th service; c_type i Indicates the computing power requirement type of the i-th service; t_d i represents the maximum tolerable delay of the ith service; c_q i Indicates the amount of data that the i-th service needs to calculate.

[0039] The technical effects of the above technical solution include: by defining a unified service description tuple s i , supports information interaction and processing between system modules; through clear time tolerance t_d i and the amount of data c_qi Provide key constraint parameters for the offloading strategy, ensuring the feasibility of decision-making; through the time-sensitive type t_type i and the computing power demand type c_type i achieve rapid classification of services, supporting hierarchical processing and resource pre-allocation of the system.

[0040] A computing offloading method for vehicle networking services for achieving the second object of the present invention includes:

[0041] Perform multi-dimensional requirement analysis on vehicle networking services to obtain the time-sensitive requirement type and computing power demand type of each vehicle networking service;

[0042] Decompose the vehicle networking services into multiple independently executable subtasks according to the execution process of each vehicle networking service; and determine the execution sequence, computing power demand, and computing resources of each subtask according to the time-sensitive requirement type and computing power demand type of the vehicle networking service.

[0043] Determine a distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type of the vehicle networking service; each level represents a different computing power demand type;

[0044] Taking the service delay and execution energy consumption as the optimization objectives, calculate the optimal computing offloading scheme under the distributed computing power fusion architecture by using the grey wolf algorithm based on the simulated annealing mechanism according to the computing power demand type of each vehicle networking service, the execution time of the vehicle networking service determined based on the execution sequence of the subtasks, and the execution energy consumption of the vehicle networking service determined based on the computing power demand and computing resources of the subtasks; the optimal computing offloading scheme realizes real-time scheduling of multi-level computing power resources by using the distributed computing power fusion architecture.

[0045] A non-transitory computer-readable storage medium for achieving the third object of the present invention, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the computing offloading method for vehicle networking services are realized.

[0046] A computer program product for achieving the fourth object of the present invention, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the computing offloading method for vehicle networking services are realized.

[0047] The beneficial effects of the present invention include:

[0048] The present invention realizes accurate characterization of vehicle networking services through dual classification of time-sensitive type and computing power demand type, improving the adaptability to complex service scenarios;

[0049] The multi-level computing power deployment of the terminal device, edge device, and cloud service can dynamically match the resource requirements of different vehicle networking tasks, optimize resource utilization, and reduce the single-point load pressure;

[0050] Taking service latency and execution energy consumption as the optimization goals, a global optimal offloading decision is achieved through an improved simulated annealing grey wolf algorithm, realizing the balance between real-time performance and energy efficiency;

[0051] The distributed architecture based on real-time computing power requirements can support the dynamic task scheduling of vehicle networking services and can well adapt to the highly dynamic in-vehicle environment. Description of the Drawings

[0052] Figure 1 It is a framework diagram of the computing offloading system for vehicle networking services in an embodiment of the present invention;

[0053] Figure 2 It is a schematic diagram of the distributed computing power fusion architecture of the three-level computing power deployment in an embodiment of the present invention;

[0054] Figure 3 It is a flowchart of the improved grey wolf algorithm integrating the simulated annealing mechanism in an embodiment of the present invention. Detailed Implementation Manner

[0055] The following detailed implementation manners are used to explain the technical solution of the present invention so that those skilled in the art can understand the present invention. The protection scope of the present invention is not limited to the following specific implementation structures. Those made by those skilled in the art that include the technical solution of the present invention and are different from the following detailed implementation manners are also within the protection scope of the present invention.

[0056] Refer to Figures 1 - 3 , this embodiment provides a computing offloading method for vehicle networking services, including the following steps:

[0057] 1. Analysis of vehicle networking service requirements

[0058] For different vehicle networking services, starting from the requirements, deeply analyze the vehicle networking services, and extract their key requirements. The key requirements include data-related requirements (such as data volume, data complexity), computing-related requirements (such as computing complexity, computing speed requirements), and network-related requirements (such as network bandwidth, network stability and reliability); classify the vehicle networking services according to the degree of time sensitivity, and divide them into three types of time-sensitive demand types: extremely time-sensitive (H), moderately time-sensitive (M), and not time-sensitive (L). The extremely time-sensitive service (H) has extremely high requirements for response time, and even a slight delay may lead to serious consequences, such as the emergency braking decision service in autonomous driving; the moderately time-sensitive service has certain requirements for response time, but allows relatively short delays. For example, the real-time traffic navigation service, a short delay (such as a few seconds) will not have a significant impact on the navigation result; the not time-sensitive service has lower requirements for response time, such as the software update service of the vehicle; and according to the key requirements, divide the computing power demand types of the services into three categories: data-intensive (D), computing-intensive (C), and network-intensive (N); for vehicle networking services with high requirements in data-related aspects, such as large data volume and high data complexity, and the key task is to process and manage a large amount of complex data, and have high requirements for the ability of data storage, reading, and processing, the corresponding computing power demand type is defined as data-intensive (D); for vehicle networking services with high computing complexity and high requirements for computing speed, and the key emphasis is on quickly completing complex computing tasks through powerful computing capabilities, the corresponding computing power demand type is defined as computing-intensive (C); the key requirements focus on network-related aspects, have high requirements for network bandwidth, and have strict standards for network stability and reliability. During operation, the vehicle networking services that rely on a stable and high-speed network connection to transmit data and implement service functions have the corresponding computing power demand type defined as network-intensive (N).

[0059] Assume that each service is represented by s i According to the service category, request service time, location and other relevant data of the vehicle networking service, each service is represented by a six-dimensional array, that is, s i =(t i ,q i ,t_type i ,c_type i ,t_d i ,c_q i ), where t i represents the request service time of the i-th service, q i represents the request service initiating device of the i-th service, such as a certain moving vehicle; t_type iIndicates the time - sensitivity type of the \(i\) - th service, and its values are \(H\) (time - extremely - sensitive service), \(M\) (time - moderately - sensitive service), or \(L\) (time - insensitive service); \(c\_type\) i Indicates the computing power demand type of the \(i\) - th service, and its values are \(D\) (data - intensive), \(C\) (computation - intensive), or \(N\) (network - intensive); \(t\_d\) i Indicates the maximum tolerable latency of the \(i\) - th service; \(c\_q\) i Indicates the amount of data that the \(i\) - th service needs to compute.

[0060] 2. Service business process construction

[0061] According to the inflow and outflow of data streams during the execution of each service, determine the main execution process of the service; decompose each vehicle - to - everything (V2X) service according to the main execution process of the service, refine it into multiple independently - executed subtasks, and determine information such as the execution sequence, computing power requirements, and computing resource constraints of each subtask according to the key requirements of each service analyzed in step 1; for example, for time - extremely - sensitive (H) V2X services, when determining the execution sequence of subtasks, prioritize the rapid execution of key subtasks to reduce latency; for computation - intensive (C) V2X services, ensure the reasonable allocation of computing resources and computing coordination between subtasks; for V2X services with high computing complexity, allocate subtasks to appropriate computing power nodes according to computing capabilities.

[0062] The maximum tolerable latency \(t\) of each V2X service di As a reference for the time constraints of subtasks, help determine the execution time range of subtasks; the amount of data to be computed \(c\_q\) i Reasonably allocate among subtasks to ensure that the amount of computation for each subtask is within a reasonable range.

[0063] Each V2X service is represented by a directed acyclic graph (\(V\) i , \(E\_edge\) i ), where \(V\) i Represents the set of subtasks of the \(i\) - th V2X service \(s\) i , \(V\) i =\(\{v\) i,1 , \(v\) i,2 , \(\cdots\), \(v\) i,n-1 , \(v\) i,n , \(\}\), \(E\_edge\) i =\(\{e\) i |e i =(v i,m , \(v\) i,n )\}\) represents the set of edges, where the \(i\) - th edge points from vertex \(v\) i,m to \(v\) i,n , indicating that subtask \(v\) i,n must be executed before subtask \(v\)i,m Execute later.

[0064] 3. Construction of a Distributed Computing Power Fusion Architecture with Three-Level Computing Power Deployment

[0065] As Figure 2 shown, construct a distributed computing power fusion architecture based on three-level computing power deployment. The bottom layer is the first computing power node network composed of terminal devices (including terminals such as vehicles and cameras), which is responsible for initiating service requests. Assume there are M terminal devices in the system; the middle layer is the second computing power node network composed of edge devices (including roadside units, base stations, etc.), which is responsible for executing lightweight tasks and providing intelligent services. Assume there are N edge devices in the system; the farthest layer is the cloud computing power node network provided by the cloud data center, which is responsible for data storage, big data mining, and sharing, providing large-scale services, and realizing the computing power fusion of three-layer heterogeneous resources at the edge and cloud. It can be seen that there are a total of M + N + 1 available computing powers in the system. The representation of the jth computing power is set as r j =(r_type j , s j , c j , n j , pc j , pt j ), where r_type j represents the computing power type of the jth computing power, including terminal (represented by s1~s M ), edge computing power (represented by s M+1 ~s M+N ), and cloud computing power (represented by s M+N+1 ); s j represents the vehicle networking service requested by the jth computing power; c j represents the computing power of the jth computing power, n j represents the network transmission capacity of the jth computing power, pc j represents the computing power of the jth computing power, and pt j represents the transmission power of the jth computing power.

[0066] 4. Generate the Optimal Offloading Scheme

[0067] Propose an improved grey wolf algorithm integrating the simulated annealing mechanism to find the optimal offloading scheme and output the optimal offloading decision. Define (s i , r_type j ) as the position of the grey wolf, that is, the offloading decision; for example, when there are 10 vehicle networking services, the position of each grey wolf is defined as follows: {(s1, r_type3), (s2, r_type1), (s3, r_type2),…, (s 10 , r_type5)}; where (s 10, (r_type5) indicates that the 10th vehicle networking service is offloaded to the 5th computing power, which may be the computing power of a terminal device, an edge device, or cloud computing power, and is specifically set according to the distributed computing power fusion architecture described in the previous step 3.

[0068] When calculating the execution time and execution energy consumption of each vehicle networking service s i , it is necessary to consider its decomposed subtasks, and the total execution time T i , E_edge i ) of the entire service s i needs to be calculated according to the order of the edges in the directed acyclic graph (V i and the total execution energy consumption E i . The total execution time T i is the accumulation of the execution times of all subtasks, and the dependency relationship between subtasks needs to be considered (that is, the order represented by the edges. Only after the previous subtasks are completed can the subsequent subtasks start to execute). The total execution energy consumption E i is the accumulation of the execution energy consumptions of all subtasks.

[0069] For each subtask v i of the service s i,j , if the computing power r_type i is the computing power of a certain terminal device in the first computing power node network s1~s M , then the execution time and execution energy consumption of the subtask v i,j are respectively: T i,j = c_q i,j / c i , E i,j = c_q i,j × pc i / c i ; if the computing power r_type i is the computing power of a certain edge device in the second computing power node network s M+1 ~s M+N , then the execution time and execution energy consumption of the subtask v i,j are respectively: T i,j = c_q i,j / c i + c_q i,j / n i , E i,j = c_q i,j × pc i / c i + c_q i,j × pt i / n i ; if r_type i is cloud computing power, then for the service s iThe execution time and execution energy consumption are respectively: T i,j = c_q i,j / c i + c_q i,j / n i , E i,j = c_q i,j × pc i / c i + c_q i,j × pt i / n i .

[0070] At this time, the total execution time and total execution energy consumption of each task corresponding to the position of each gray wolf can be calculated as: ∑T i and ∑E i ; n is the number of subtasks of service S i . Therefore, the objective function value of the optimal gray wolf can be defined as f(s i , r_type j ) = ∑T i / ∑t_d i + ∑E i / ∑e_d i , where ∑t_d i represents the maximum allowable execution time of all current Internet of Vehicles tasks; ∑e_d i represents the system energy consumption of all current Internet of Vehicles tasks under the maximum time delay. If ∑T i > ∑t_d i , it means that this offloading decision does not meet the time constraint requirements, then f(s i , r_type j ) = 0.

[0071] The algorithm process is as Figure 3 shown. First, all parameters of the algorithm are initialized, including the population size, the maximum number of iterations h, the initial temperature value, etc., and the positions of the individuals in the gray wolf population are randomly initialized. The objective function value of the gray wolf group is calculated, and the optimal position is recorded. Secondly, the positions of the gray wolves are updated according to the simulated annealing mechanism. To avoid falling into a local optimal solution, the probability of accepting a new position can be defined as Perform position screening, where f and f' are the objective function values of the current position and the new position respectively, and T(h) represents the temperature parameter in the simulated annealing process, which is a function of the iteration number h. If p > the set value (0.50 in this embodiment), the new position is accepted; otherwise, it is not accepted. Determine whether the iteration condition is satisfied. If it is satisfied, output the optimal solution, which is the optimal computing offloading scheme. The optimal offloading scheme is an optimization strategy that comprehensively considers various factors and aims to improve the resource utilization rate, service quality of the vehicle network system, reduce energy consumption and cost through reasonable computing task offloading, so as to achieve the optimal overall performance of the system. The improved grey wolf algorithm integrating the simulated annealing mechanism is to efficiently search for this optimal scheme among numerous possible offloading schemes. For example, simple data processing tasks can be offloaded to the in-vehicle computing unit, and complex deep learning model inference tasks are assigned to the cloud server to avoid overloading some devices while leaving other devices idle, improving the utilization rate of the overall computing power resources. For example, when the network bandwidth is sufficient, offload the computing tasks of a large amount of data to the cloud, and in areas with weak network signals, preferentially select local or nearby edge servers for computing. For example, preferentially select edge computing nodes that are close to the vehicle and have fast computing speed for offloading. For example, offload important vehicle diagnostic services to multiple different computing nodes to ensure that service results can still be obtained normally when a certain node fails. For example, offload some tasks to external computing resources to reduce the energy consumption of local devices and extend the battery life of the devices. For example, for some periodic data processing tasks, they can be offloaded to the cloud for processing when the vehicle is connected to an external power source or the network condition is good.

[0072] 5. Complete real-time computing power scheduling.

[0073] According to the optimal computing offloading scheme, realize the real-time and efficient scheduling of three-level computing power resources, and improve the service quality of the vehicle network, including service delay and system energy consumption.

[0074] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0075] The embodiment of the present invention also provides a computing offloading system for vehicle network services, including:

[0076] Service demand analysis module: used to perform multi-dimensional demand analysis on vehicle network services to obtain the time-sensitive demand type and computing power demand type of each vehicle network service;

[0077] Service process construction module: used to decompose vehicle networking services into multiple independently executable subtasks according to the execution process of each vehicle networking service; and determine the execution sequence, computing power requirements, and computing resources of each subtask according to the time-sensitive demand type and computing power demand type of the vehicle networking service.

[0078] Computing power fusion module: used to determine a distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type of the vehicle networking service; each level represents a different computing power demand type.

[0079] Computing offloading module: used to calculate the optimal computing offloading scheme under the distributed computing power fusion architecture by taking service latency and execution energy consumption as optimization objectives, using the Grey Wolf algorithm based on the simulated annealing mechanism according to the computing power demand type of each vehicle networking service, the execution time of the vehicle networking service determined based on the execution sequence of subtasks, and the execution energy consumption of the vehicle networking service determined based on the computing power requirements and computing resources of subtasks; the optimal computing offloading scheme realizes real-time scheduling of multi-level computing power resources by using the distributed computing power fusion architecture. The expression of the optimal computing offloading scheme is: {(s1, r_type a,1 ), (s2, r_type a,2 ), …, (s n-1 , r_type a,n-1 ), (s n , r_type a,n )}, (s2, r_type a,2 ) means that the vehicle networking service of the second computing power request is offloaded to the computing power type numbered (a, 2), and the computing power type includes terminal devices, edge devices, or cloud computing power.

[0080] In some embodiments, in the service demand analysis module, the method for obtaining the time-sensitive demand type and computing power demand type of each vehicle networking service includes:

[0081] Analyze each vehicle networking service according to the data flow, computing process, and network interaction dimensions, and extract key requirements; the key requirements include: data-related requirements, computing-related requirements, and network-related requirements; data-related requirements include data volume and data complexity; computing-related requirements include computing complexity and computing speed requirements; network-related requirements include network bandwidth, network stability, and reliability.

[0082] Determine the time-sensitive demand type of each vehicle networking service according to the network-related requirements; the time-sensitive demand type includes time extremely sensitive (H), time moderately sensitive (M), and time insensitive (L).

[0083] Determine the computing power demand type of each vehicle networking service according to data-related requirements and computing-related requirements. The computing power demand types include data-intensive (D), computing-intensive (C), and network-intensive (N).

[0084] In some embodiments, the method for decomposing a vehicle networking service into multiple independently executed subtasks includes:

[0085] For time-critical (H) services: Decompose the service into a strictly serially executed subtask chain;

[0086] For moderately time-sensitive (M) services: Decompose into a subtask sequence allowing short-delay elastic scheduling, supporting dynamic merging and delayed execution of subtasks;

[0087] For time-insensitive (L) services: Decompose into a set of asynchronous batch-processing subtasks, supporting subtask interruption and idle resource scheduling.

[0088] In some embodiments, each vehicle networking service is represented by a directed acyclic graph (V i , E_edge i ), where V i represents the set of subtasks of the i-th vehicle networking service s i , V i = {v i,1 , v i,2 , …, v i,n-1 , v i,n}, and E_edge i = {e i | e i = (v i,m , v i,n )} represents the set of edges, where the i-th edge points from vertex v i,m to v i,n , indicating that subtask v i,n must be executed after subtask v i,m .

[0089] In some embodiments, the method for determining the execution time and execution energy consumption of each vehicle networking service includes:

[0090] When the computing power demand type of the vehicle networking service is provided by the first computing power node network composed of terminal devices, the execution time and execution energy consumption of each subtask of the vehicle networking service are respectively: T i,j = c_q i,j / c i , E i,j = c_q i,j × pc i / c i ;

[0091] When the computing power demand type of the vehicle networking service is provided by the second computing power node network composed of edge devices, the execution time and execution energy consumption of each subtask of the vehicle networking service are respectively: T i,j = c_q i,j / c i + c_q i,j / n i ;

[0092] When the computing power demand type of the vehicle networking service is provided by the cloud computing power node network, the execution time and execution energy consumption of each subtask of the vehicle networking service are respectively: T i,j = c_q i,j / c i + c_q i,j / n i , E i,j = c_q i,j × pc i / c i + c_q i,j × pt i / n i ;

[0093] T i,j and E i,j respectively represent the execution time and execution energy consumption of the j-th subtask of the i-th vehicle networking service; c_q i,j represents the amount of data that needs to be calculated for the j-th subtask of the i-th vehicle networking service; c i represents the computing power of the computing power type to which the i-th vehicle networking service belongs; n i represents the network transmission capacity of the computing power type to which the i-th vehicle networking service belongs; pc i represents the computing power of the computing power type to which the i-th vehicle networking service belongs; pt i represents the transmission power of the computing power type to which the i-th vehicle networking service belongs;

[0094] Adding up the execution times of all subtasks, the execution time of the vehicle networking service is obtained;

[0095] Adding up the execution energy consumptions of all subtasks, the execution energy consumption of the vehicle networking service is obtained.

[0096] In some embodiments, the calculation method of the optimal computing offloading scheme includes:

[0097] Setting the position of each gray wolf as (s i , r_type j ), where s i represents the vehicle networking service, and r_type j represents the computing power type assigned to the vehicle networking service;

[0098] Set the objective function value of the optimal gray wolf to be f(s i ,r_type j )=∑T i / ∑t_d i +∑E i / ∑e_d i ; Among them, ∑t_d i Indicates the maximum allowed execution time of all current Internet of Vehicles tasks; ∑e_d i Indicates the system energy consumption of all current Internet of Vehicles tasks under the maximum delay condition;

[0099] Initialize all parameters of the simulated annealing algorithm, including population size, maximum number of iterations, initial temperature value, and randomly initialize the positions of individuals in the gray wolf population, calculate the objective function value of the gray wolf population, and record the optimal position;

[0100] Update the gray wolf's position according to the simulated annealing mechanism, and define the probability of accepting the new position as The position is selected, where f and f' are the objective function values ​​of the current position and the new position respectively, and T(h) represents the temperature parameter in the simulated annealing process; if p is greater than the set value, the gray wolf position is updated to the new position;

[0101] Determine whether the iteration condition is met, if not, repeat the above steps of updating the gray wolf position; if met, output the optimal solution, which is the optimal computation offloading solution.

[0102] In some embodiments, the Internet of Vehicles service i Includes: i =(t i ,q i ,t_type i ,c_type i ,t_d i ,c_q i );t i represents the request service time of the i-th service, q i Indicates the service initiator device requesting the i-th service; t_type i Indicates the time-sensitive type of the i-th service; c_type i Indicates the computing power requirement type of the i-th service; t_d i represents the maximum tolerable delay of the ith service; c_q i Indicates the amount of data that the i-th service needs to calculate.

[0103] An embodiment of the present invention further provides a non-transitory computer-readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, each step of the method described in the present invention is implemented, which will not be elaborated here.

[0104] The computer-readable storage medium may be an internal storage unit of the data transmission device or computer device provided in any of the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device.

[0105] Furthermore, the computer-readable storage medium may also include both the internal storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store the data to be output or already output.

[0106] Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system, or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0108] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more processes and / or blocks Figure 1 in the process Figure 1 or processes and / or blocks

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more processes and / or blocks Figure 1 in the process Figure 1 or processes and / or blocks

[0110] An embodiment of the present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the computing offloading method for vehicle networking services

[0111] Content not described in detail in this specification belongs to the prior art well known to those skilled in the art

Claims

1. A computing offloading system for Internet of Vehicles services, characterized in that: include: Service demand analysis module: used to perform multi-dimensional demand analysis on Internet of Vehicles services and obtain the time-sensitive demand type and computing power demand type of each Internet of Vehicles service; Service process building module: used to decompose the business of the Internet of Vehicles service according to the execution process of each Internet of Vehicles service to obtain multiple independently executed subtasks; and determine the execution sequence, computing power requirements, and computing resources of each subtask according to the time-sensitive demand type and computing power demand type of the Internet of Vehicles service; Computing power fusion module: used to determine the distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type of the Internet of Vehicles service; each level represents a different computing power demand type; Computing offloading module: It is used to calculate the optimal computing offloading solution under the distributed computing power fusion architecture based on the computing power requirement type of each Internet of Vehicles service, the execution time of the Internet of Vehicles service determined based on the order of subtask execution, and the execution energy consumption of the Internet of Vehicles service determined based on the computing power requirements and computing resources of the subtasks, with service delay and execution energy consumption as optimization goals, and adopts the gray wolf algorithm based on the simulated annealing mechanism; the optimal computing offloading solution utilizes the distributed computing power fusion architecture to realize real-time scheduling of multi-level computing power resources.

2. The computing offloading system for Internet of Vehicles services according to claim 1, characterized in that: In the service demand analysis module, the method for obtaining the time-sensitive demand type and computing power demand type of each Internet of Vehicles service includes: Analyze each Internet of Vehicles service based on data flow, computing process and network interaction dimensions to extract key requirements; the key requirements include: data-related requirements, computing-related requirements and network-related requirements; Determine the time-sensitive demand type for each connected vehicle service based on network-related requirements; Determine the type of computing power required for each Internet of Vehicles service based on data-related and computing-related requirements.

3. The computing offloading system for Internet of Vehicles services according to claim 1, characterized in that: The time-sensitive demand types include particularly time-sensitive services, medium time-sensitive services and time-insensitive services; Methods for decomposing the Internet of Vehicles service into multiple independently executed subtasks include: For time-sensitive services: break down the service into multiple subtasks that are executed serially; For medium-time-sensitive services: break them down into multiple subtasks that allow flexible scheduling; For time-insensitive services: break them down into multiple subtasks for asynchronous batch processing.

4. The computing offloading system for Internet of Vehicles services as claimed in claim 3, characterized in that: Each Internet of Vehicles service uses a directed acyclic graph (V i ,E_edge i ) indicates that V i represents the i-th Internet of Vehicles service s i The subtask set, V i = {v i,1 , v i,2 , …, v i,n-1 , v i,n ,},E_edge i ={e i |e i =(v i,m ,v i,n )} represents a set of edges, where the i-th edge consists of vertex v i,m Point to v i,n , represents the subtask v i,n Subtask v i,m Then execute.

5. The computing offloading system for Internet of Vehicles services according to claim 1, characterized in that: Methods for determining the execution time and execution energy consumption of each connected vehicle service include: When the computing power requirement type of the Internet of Vehicles service is provided by the first computing power node network composed of terminal devices, the execution time and execution energy consumption of each subtask of the Internet of Vehicles service are respectively: T i,j =c_q i,j / c i , E i,j =c_q i,j ×pc i / c i ; When the computing power requirement type of the Internet of Vehicles service is provided by the second computing power node network composed of edge devices, the execution time and execution energy consumption of each subtask of the Internet of Vehicles service are respectively: T i,j =c_q i,j / c i +c_q i,j / n i ; When the computing power requirement type of the Internet of Vehicles service is provided by the cloud computing power node network, the execution time and execution energy consumption of each subtask of the Internet of Vehicles service are respectively: T i,j =c_q i,j / c i +c_q i,j / n i ,E i,j =c_q i,j ×pc i / c i +c_q i,j ×pt i / n i ; T i,j and E i,j They represent the execution time and energy consumption of the jth subtask of the i-th connected vehicle service respectively; c_q i,j represents the amount of data that needs to be calculated for the jth subtask of the i-th Internet of Vehicles service; c i Indicates the computing power of the computing power type to which the i-th Internet of Vehicles service belongs; n i represents the network transmission capacity of the computing power type to which the i-th Internet of Vehicles service belongs; pc i Indicates the computing power of the computing power type to which the i-th Internet of Vehicles service belongs; pt i Indicates the transmission power of the computing power type to which the i-th Internet of Vehicles service belongs; Add up the execution time of all subtasks to get the execution time of the Internet of Vehicles service; The execution energy consumption of all subtasks is added together to obtain the execution energy consumption of the Internet of Vehicles service.

6. The computing offloading system for Internet of Vehicles services according to claim 1, characterized in that: The calculation method of the optimal computation offloading solution includes: Set the position of each gray wolf to (s i ,r_type j ), where s i Indicates Internet of Vehicles services, r_type j Indicates the type of computing power allocated to the Internet of Vehicles service; Set the objective function value of the optimal gray wolf to be f(s i ,r_type j )=∑T i / ∑t_d i +∑E i / ∑e_d i ; Among them, ∑t_d i Indicates the maximum allowed execution time of all current Internet of Vehicles tasks; ∑e_d i Indicates the system energy consumption of all current Internet of Vehicles tasks under the maximum delay condition; Initialize all parameters of the simulated annealing algorithm, including population size, maximum number of iterations, initial temperature value, and randomly initialize the positions of individuals in the gray wolf population, calculate the objective function value of the gray wolf population, and record the optimal position; Update the gray wolf's position according to the simulated annealing mechanism, and define the probability of accepting the new position as The position is selected, where f and f' are the objective function values ​​of the current position and the new position respectively, and T(h) represents the temperature parameter in the simulated annealing process; if p is greater than the set value, the gray wolf position is updated to the new position; Determine whether the iteration condition is met, if not, repeat the above steps of updating the gray wolf position; if met, output the optimal solution, which is the optimal computation offloading solution.

7. The computing offloading system for Internet of Vehicles services according to claim 6, characterized in that: The Internet of Vehicles Services i Includes: i =(t i ,q i ,t_type i ,c_type i ,t_d i ,c_q i );t i represents the request service time of the i-th service, q i Indicates the service initiator device requesting the i-th service; t_type i Indicates the time-sensitive type of the i-th service; c_type i Indicates the computing power requirement type of the i-th service; t_d i represents the maximum tolerable delay of the ith service; c_q i Indicates the amount of data that the i-th service needs to calculate.

8. A method for offloading computing for Internet of Vehicles services of the system as claimed in claim 1, characterized in that: include: Perform multi-dimensional demand analysis on Internet of Vehicles services to obtain the time-sensitive demand type and computing power demand type of each Internet of Vehicles service; Decomposing the Internet of Vehicles service into multiple independently executed subtasks according to the execution process of each Internet of Vehicles service; And determine the execution order, computing power requirements, and computing resources of each subtask based on the time-sensitive demand type and computing power demand type of the Internet of Vehicles service; Determine a distributed computing power fusion architecture based on multi-level computing power deployment according to the computing power demand type of the Internet of Vehicles service; each level represents a different computing power demand type; According to the computing power requirement type of each Internet of Vehicles service, the execution time of the Internet of Vehicles service determined based on the order of subtask execution, and the execution energy consumption of the Internet of Vehicles service determined based on the computing power requirements and computing resources of the subtasks, and with service delay and execution energy consumption as optimization goals, the gray wolf algorithm based on the simulated annealing mechanism is used to calculate the optimal computing offloading solution under the distributed computing power fusion architecture; the optimal computing offloading solution utilizes the distributed computing power fusion architecture to realize real-time scheduling of multi-level computing power resources.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the computing offloading method for Internet of Vehicles services as claimed in claim 8 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the computing offloading method for Internet of Vehicles services described in claim 8 are implemented.

Images