Vehicle edge computing predictive task offloading method, system, and storage medium

Abstract

The present application belongs to the technical field of Internet of Vehicles, and particularly relates to a vehicle edge computing predictive task offloading method, a system and a storage medium. The digital twin assisted vehicle edge computing predictive task offloading method comprises the following steps: perceiving all task offloading demands in a domain; receiving available computing resource states from edge vehicles and forwarding the same to a digital twin module; performing parameter simulation through the digital twin module, judging whether a fog node accepts the task offloading demand through a parameter simulation result and an actual resource inventory, and obtaining a task offloading decision; for the accepted task offloading decision, the edge vehicle offloads a task to the fog node for task execution, and then sends an output result back to the edge vehicle. The present application further provides a system for implementing the above method. The present application can reduce task completion delay, improve fog node resource allocation balance, adapt to virtual-real error and improve VEC system computing rate, and has a good application prospect.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of Internet of Vehicles, and particularly relates to a vehicle edge computing predictive task offloading method, system and storage medium. BACKGROUND

[0002] The progress of Internet of Things and wireless communication technology has driven the explosive growth of intelligent terminal devices and data traffic. It is estimated that in the 6G era, millions of terminal devices will access wireless networks, and mobile applications such as network games, virtual / augmented reality and autonomous driving will generate a large amount of data at the edge of wireless networks. These data are transmitted to the core network through technologies such as wireless communication, GPS and sensors, bringing unprecedented challenges to the existing wireless network architecture.

[0003] Edge computing effectively improves the performance of services and applications by migrating computing and caching functions to the network edge. However, due to the limited resources of a single edge cloud or terminal device, multiple edge nodes or core clouds need to be coordinated to complete tasks. Vehicle edge computing (VEC) network is a concept that combines mobile edge computing (MEC) and Internet of Vehicles (IoV) technology, which solves the problem of limited computing power of vehicle terminals by sinking vehicle tasks to the network edge. VEC network aims to reduce the time delay and energy consumption of executing terminal tasks, and improve the quality of service and user experience.

[0004] In recent years, digital twin, as an important means of physical and information fusion, has attracted widespread attention from various industries due to its characteristics of synchronous mapping of virtual and real, symbiosis and integration, and has provided a feasible solution to solve the dynamic complexity of VEC network. Digital twin establishes accurate mapping with VEC network, monitors its dynamic changes in real time, captures the characteristics of service demand and resource utilization, and realizes the function of efficient auxiliary intelligent decision-making. Currently, there have been attempts to do so, such as: “IEEE Transactions on Intelligent Vehicles, 2022, 2(3): 212-218.”, “IEEE Internet of Things Journal, 2022, 6(04): 125-133.”, “IEEE Transactions on Industrial Informatics, 2021, 18(2): 1405-1413.”, “IEEE Communications Magazine, 2019, 57(5): 64-69.”, etc.

[0005] Although some research has been done on simulation applications based on digital twinning, the dynamic mapping of simulation models still needs to be improved. In the actual environment of vehicle edge computing, digital twinning technology provides the ability to dynamically update simulation models, which can respond to disturbances during operation and maintain consistency between the virtual and real worlds. However, in practical engineering applications, in order to meet the high requirements of algorithms on the diversity of disturbance types and identification accuracy, complex algorithms and a large amount of running data are usually required. Digital twinning systems need to address the above issues to ensure the accuracy and timeliness of simulation results. Therefore, there is an urgent need in the field to develop new task offloading strategies to more reasonably select tasks to be executed by terminal devices, edge computing servers or cloud computing servers in real time, minimize system latency and resource consumption, and achieve rationalization of wireless network resource consumption and node resource allocation. SUMMARY

[0006] To solve the problems of the prior art, the present application provides a digital twinning assisted vehicle edge computing predictive task offloading method and system.

[0007] A digital twinning assisted vehicle edge computing predictive task offloading method, comprising the following steps performed at a fog node:

[0008] Step 1: sensing all task offloading requirements in the domain;

[0009] Step 2: receiving the available computing resource status from the edge vehicle and forwarding it to the digital twinning module;

[0010] Step 3: performing parameter simulation through the digital twinning module, determining whether the fog node accepts the task offloading requirement based on the parameter simulation results and the actual resource inventory, and obtaining a task offloading decision;

[0011] Step 4: for the accepted task offloading decision, the edge vehicle offloads the task to the fog node for task execution, and then sends the output result back to the edge vehicle.

[0012] Preferably, in step 1, the fog node is responsible for maintaining a list of edge vehicles accessing and leaving within the communicable network range of the fog node; the edge vehicle is initialized when it accesses and sends its task offloading requirement to the fog node.

[0013] Preferably, in step 3, the digital twinning module is configured to create the network topology of all connected nodes and the task interaction relationship, use the digital twinning model to perform parameter simulation on the task offloading decision and data transmission method of the actual fog node or edge node, and realize prediction of the task computing capability of the fog node.

[0014] Preferably, step 3 specifically comprises the following steps:

[0015] Step 3.1: inputting the historical task offloading data volume generated in the domain into the integrated artificial intelligence model of the digital twin model;

[0016] Step 3.2: calculating the computation time required for executing the next-time vehicle task in combination with the vehicle movement mode, available resource state and the historical task offloading data volume;

[0017] Step 3.3: comparing the computation time of the next-time vehicle task with the task-waitable delay contained in the task offloading demand, and if the computation time of the next-time vehicle task is lower than the task-waitable delay, the fog node accepts the task offloading demand, otherwise, it rejects.

[0018] Preferably, in step 3.1, the artificial intelligence model algorithm is selected from LSTM, RNN, CNN, Bayesian method or ARIMA time series model.

[0019] Preferably, in step 3.3, the standard for judging whether the fog node accepts the task offloading decision further includes that the edge vehicle is directly connected with the fog node, and the fog node can accept the task offloading decision when executing the task, otherwise, it rejects.

[0020] Preferably, step 4 specifically includes the following steps:

[0021] Step 4.1: when the task offloading decision is accepted, the edge vehicle offloads the task to the fog node, and all accepted tasks are saved in the local queue of the fog node until the resource meets the demand for stably executing the task;

[0022] Step 4.2: after the task is executed, all output results are stored in the result queue of the fog node;

[0023] Step 4.3: the scheduling program of the fog node sends the output results back to the edge vehicle.

[0024] Preferably, the following steps are executed locally on the edge vehicle:

[0025] Step a: the edge vehicle generates a task and stores the task in the local queue;

[0026] Step b: the resource manager of the edge vehicle perceives the task offloading decision basis in real time, calculates the demand of the task and the current work load of the device, decides whether the task is processed locally or by the fog node, and for the task processed by the fog node, obtains the task offloading demand; the task offloading decision basis includes vehicle metadata, task data, fog resource and task deadline information;

[0027] Step c: temporarily storing the task offloading demand in the offloading queue;

[0028] Step d: when the edge vehicle is connected with the fog node, sending the task offloading demand.

[0029] The application also provides a system for implementing the above-mentioned digital twin assisted vehicle edge computing predictive task offloading method, comprising: a fog node and at least one edge vehicle; the fog node is connected with a digital twin module.

[0030] The application also provides a computer readable storage medium, which stores a computer program for implementing the above-mentioned digital twin assisted vehicle edge computing predictive task offloading method, or a computer program for implementing the above-mentioned system.

[0031] In the application, the "task" is data that needs to be calculated; the "task offloading" is to process the data that needs to be calculated in the fog node or locally; and the "task offloading decision" is the decision result of processing the data that needs to be calculated in the fog or locally.

[0032] The application provides a digital twin assisted vehicle edge computing predictive task offloading framework, realizes digital twin mapping between edge vehicles and fog nodes, estimates, predicts and evaluates the real-time state of the VEC network by creating a virtual copy, provides a method for allocating node resources according to the digital twin parameter prediction result, can optimize the utilization rate of edge devices, minimize the task completion delay and realize balanced task offloading between fog nodes. Through simulation verification, it is found that the scheme proposed in the present application is superior to other existing frameworks in reducing the task completion delay, improving the balance of fog node resource allocation, adapting to virtual-real errors and improving the calculation rate of the VEC system. Therefore, the application has good application prospect.

[0033] Obviously, according to the above content of the application, according to the ordinary technical knowledge and common means in the art, other various forms of modifications, replacements or changes can be made without departing from the above-mentioned basic technical ideas of the application.

[0034] The above content of the application will be further described in detail through the specific embodiments in the form of examples. However, this should not be understood as limiting the scope of the above-mentioned subject matter of the application to the following examples. Any technology realized based on the above-mentioned content of the application belongs to the scope of the application. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 It is a schematic diagram of the system of the application. DETAILED DESCRIPTION

[0036] It should be particularly noted that the algorithms of data acquisition, transmission, storage and processing steps not specifically described in the examples, and the hardware structure, circuit connection and the like not specifically described can be realized through the existing technology disclosed.

[0037] Example 1: Digital Twin-Assisted Vehicle Edge Computing Predictive Task Offloading Method and System

[0038] This embodiment provides a vehicle edge computing (VEC) network capable of executing a digital twin-assisted vehicle edge computing predictive task offloading method.

[0039] Specifically, the system is as follows Figure 1 As shown, the system includes interconnected fog nodes and edge vehicles (autonomous vehicles). Edge vehicles collect data via sensors, upload it to the fog nodes, and use the data to create virtual replicas to estimate, predict, and evaluate the real-time state of the VEC (Vehicle Controller). Each fog node integrates a digital twin module. This digital twin module is configured to: create the network topology of all connected nodes (including node speed and direction, road topology) and task interaction relationships (including task timeliness information, data scale, and computation time); and use the digital twin model to perform parameter simulation (including resources, task deadlines, vehicle mobility, network topology, and other parameters) of the task offloading decisions and data transmission methods of the actual fog nodes or edge nodes, thereby predicting the task computing capabilities of the fog nodes.

[0040] The specific steps of the method for predictive task offloading of edge computing for digital twin-assisted vehicles using the above system are as follows:

[0041] Perform the following steps locally on the edge vehicle:

[0042] Step a: Edge vehicles generate tasks and store them in a local queue;

[0043] Step b: The resource manager of the edge vehicle perceives the task unloading decision criteria in real time, calculates the task requirements and the current workload of the device, and decides whether the task is processed locally or by the fog node. For tasks processed by the fog node, the task unloading requirements are obtained. The task unloading decision criteria include vehicle metadata, task data, fog resources, and task deadline information.

[0044] Step c: Temporarily store the task unloading requirements in the unloading queue;

[0045] Step d: When the edge vehicle connects to the fog node, send a task unloading request.

[0046] Perform the following steps at the fog node:

[0047] Step 1: Sensing all task offloading requests within the domain; In this step, the fog node is responsible for maintaining a list of edge vehicles that enter and leave within the fog node's communicable network range; When an edge vehicle enters, it initializes and sends its task offloading request to the fog node.

[0048] Step 2: Receive the status of available computing resources from the edge vehicle and forward it to the digital twin module;

[0049] Step 3: parameter simulation is performed through the digital twin module, whether the fog node accepts the task offloading demand is judged through the parameter simulation result and the actual resource stock, and a task offloading decision is obtained;

[0050] Step 4: for the accepted task offloading decision, the edge vehicle offloads the task to the fog node for task execution, and then sends the output result back to the edge vehicle.

[0051] Step 3 specifically includes the following steps:

[0052] Step 3.1: input the historical task offloading data volume generated in the domain into the integrated artificial intelligence model of the digital twin model;

[0053] Step 3.2: calculate the calculation time required for executing the vehicle task at the next moment in combination with the vehicle movement mode, the available resource state and the historical task offloading data volume;

[0054] Step 3.3: judge whether the fog node accepts the task offloading decision, the criteria including:

[0055] 1) compare the calculation time of the vehicle task at the next moment with the task waitable time delay contained in the task offloading demand, if the calculation time of the vehicle task at the next moment is lower than the task waitable time delay, the fog node accepts the task offloading demand, otherwise, it is rejected;

[0056] 2) the edge vehicle is directly connected with the fog node, and the fog node can accept the task offloading decision when executing the task, otherwise, it is rejected.

[0057] The following part is the basic definition of the vehicle task offloading model, which is a modeling representation of step 3.3:

[0058] The fog / edge collaborative computing mode of the vehicle is determined by the calculation time of the vehicle offloading task at the next moment of the system And the deadline of the incoming task. After completing the task calculation in the space L, the task will make a decision to be processed by the fog node or the edge, and the task offloading mode is based on the completion time T i Make a decision, and the decision-making process can be represented by the following formula:

[0059]

[0060] Where 0 represents local execution, 1 represents that the task has timed out when being offloaded, and ∩ represents. O ij Represents the task to be executed. i and j are both serial numbers, and belong to V (vehicle node set) and R (fog node resource set) respectively.

[0061] The model introduces a term "resource usage cost", which is a comprehensive calculation and definition of the vehicle movement mode, available resource state and the historical task offloading data volume of "step 3.2". That is, the three values of movement, resource and historical data jointly affect the resource usage cost. Then, the "total cost of resource usage" is the total cost generated by the task "vehicle movement mode, available resource state and the historical task offloading data volume"; and the "partial task cost at the fog end" is the partial cost generated by the vehicle movement mode, available resource state and the historical task offloading data volume at the fog end.

[0062] The following formula illustrates how to calculate the resource usage cost using the three values of movement, resource and historical data:

[0063] According to the characteristics of the fog nodes in the fog cluster, such as high mobility, unpredictable behavior and resource limitation, the resource usage cost is calculated for the task task i In the process of resource allocation, the stability of the task task i State information includes movement information dis t , and available computing resources F t of the fog node. The movement information and computing resources can be represented as: Assuming that the historical task offloading data volume is D t ∈{D1,D2,…,D T}, D1,D2,…,D t―1 are introduced into the resource usage cost calculation model, and the task task i The total system resource usage cost calculation state space can be represented as: s t ={D t ,dis t ,F t}.

[0064] The formula is defined as follows:

[0065]

[0066] price=γ s s t

[0067] Fog_price=γ s s t +F t total

[0068] Where F i total is the total energy consumption of the fog node in the i-th period; γ s is the offloading factor, Ft cur representing the current vehicle t time remaining resources, representing the resource threshold, when the vehicle remaining resources below the threshold, will no longer be actively involved in the task offloading decision process, only to keep their own existing offloading mode, until the vehicle's remaining resources above the threshold will participate in the task offloading decision process again.

[0069] The embodiment defines the vehicle available resource cost price is 1, the fog end alone using resource cost Fog_price set to n, and then the required resource cost of fog / edge task offloading can be calculated. The required resource cost of the fog end part of the task offloading,

[0070] Pay M = n i · price

[0071] Fog end processing task remaining 1-r i partial resource cost:

[0072]

[0073] where Pay M is the total offloading cost of the fog end, Pay C Fog end processing task remaining partial resource cost, n i is the best allocation of computing resource blocks u j .

[0074] Here is a definition of a resource block, a resource block can be defined as a division unit of resource usage cost, resource usage cost can be composed of many equally divided resource blocks.

[0075] The following is the model details contained in step 3.3:

[0076] 1. The offloaded task weighted cost calculation:

[0077] The next time the vehicle offloads the task calculation time

[0078] The payment weight of resource usage is set to α, which does not exceed the maximum number of blocks that can be allocated at one time, n i is the best allocation of computing resource blocks for the current task, task i The weighted cost model is represented as:

[0079]

[0080] n i ≤ n

[0081] where β is the weight of the delay, The actual time delay for processing tasks at the fog end, The weighted cost of task task i .

[0082] 2. The waiting time delay of the offloaded task:

[0083] When the allocated resource blocks do not exceed the maximum number of allocatable resource blocks, when the remaining resource blocks are insufficient to meet the demand of the next task, the waiting time delay is required to wait for the completion of the processing of the previous task, and then the processing is performed after meeting the demand. When the resource demand is met, the waiting time delay is the maximum calculation time delay of the previous completed task, and the time delay of the subsequent task also accumulates the waiting time delay of the previous task, so the waiting time delay T wait is represented as:

[0084]

[0085] wherein, is the accumulated time delay of the fog end processing all tasks, is the time delay of the fog end processing the task.

[0086] 3. Vehicle offloading task time calculation:

[0087] The time cost required for the fog end and the edge end to perform the vehicle offloading task at the next time is calculated in combination with the vehicle movement mode and the available resource state, and then the calculation time of the vehicle offloading task is obtained:

[0088]

[0089] wherein E i represents the efficiency of the i-th fog node, W represents the corresponding work load, and the work load is limited by the resource cost capacity of the fog node at this time. λ t represents the task arrival rate. Task t is processed by the fog node or the edge node in the calculation space L with a probability of p, and ∩ represents the task deadline, which is described by , wherein l∈L ρ(l) = 1, and the actual processing time of task i needs to meet the task deadline.

[0090] 4. Vehicle task offloading determination:

[0091] The best offloading decision of task task i is to perform calculation and processing immediately after uploading, and waiting will reduce the task calculation and processing time, and the time delay weight β will be updated accordingly. If the task waiting time delay is equal to or greater than the deadline, at this time α = 0 and β = 1, so α and β satisfy the following relationship:

[0092]

[0093] The artificial intelligence model algorithm is selected from a Long Short-Term Memory (LSTM) model. The LSTM network is a recurrent neural network, and the prediction of the task-related parameters can be realized by combining the LSTM-based deep learning algorithm. The LSTM is suitable for long-term data sequence modeling, and the algorithm can effectively retain the long-term characteristics in the data, and is suitable for edge computing vehicle trajectory prediction. Assuming that the historical task data amount generated by the vehicle node is D t ∈{D1,D2,...,D T}, D1,D2,...,D T The LSTM prediction model is imported, and the task data amount at the next moment is predicted. The goal of the prediction model is to improve the task information prediction accuracy as much as possible, that is

[0094] Step 4 specifically includes the following steps:

[0095] Step 4.1: When the task offloading decision is accepted, the edge vehicle offloads the task to the fog node, and all accepted tasks are saved in the local queue of the fog node until the resources meet the demand for stable execution of the task;

[0096] Step 4.2: After the task is executed, all output results are stored in the result queue of the fog node;

[0097] Step 4.3: The scheduler of the fog node uses the first-come-first-served algorithm to send the output results back to the edge vehicle.

[0098] In this embodiment, the digital twin module integrates physical entities, sensor communication, signal conditioning, data acquisition and control systems, etc. by establishing a physical model and a three-dimensional model of the fog node. After dynamic edge vehicle characteristic analysis, the data storage of the fog node collects and stores different basic equipment and network data information.

[0099] Finally, the system can obtain four independent state data of vehicle state, fog node state, network state and wireless channel state. The digital twin module extracts data in a mapping form for calculating, defining and describing the basic characteristics of physical components and networks in the edge computing vehicle network. By utilizing the individualized adaptability of the fog end, the real-time control execution capability and the perception capability of the edge end, the module takes the fog node as the hub of service requests, and provides a dynamic consistency model with high availability in the operation of the Internet of Vehicles.

[0100] The modeling of the digital twin assisted fog node and the method and formula of resource calculation in the model are described as follows:

[0101] The digital twin assisted environment generated by the digital twin module assisted fog node can be represented as V=v1,v2,v3,...,vN , where V is an edge vehicle node, FN represents a set of fog nodes, and the mobile edge node is denoted by M = m1, m2, m3,..., m K , where q(t) represents the task queue at time t, and Q represents the length of the task queue. Assuming that L represents the distance between adjacent fog nodes, the parameters closely related to the edge vehicle node v in the digital twin scenario are represented by the following set:

[0102]

[0103] where Q v (t) represents the queue state, represents the local computing resources of the vehicle, represents the current location of the vehicle, γ(t) represents the task pattern at time t, and θ v (t) represents other parameters related to the vehicle. The task queue in the digital twin mode can be represented as The input data size of the task is denoted by d i . The processing amount required to complete the task T i is denoted by cr i (in CPU cycles). The maximum delay of this task is The task processing adopts a first-come-first-served mode. The available resources of all nodes in the set of fog nodes are denoted by FN r , R is the set of fog node resources, r represents the fog node resource number, and the internal computing capability that the resource can provide for the task i ∈ FN r is P ci . The number of available resources for each FN is n.

[0104] The specific model and parameter calculation method are as follows:

[0105] 1. Vehicle edge computing communication model

[0106] In step a, the basic model is as follows:

[0107] The time consumption of data transmission between all connected edge vehicle nodes and fog nodes is calculated, and the wired or wireless data offloading rate between two nodes is Od ij .

[0108]

[0109] where i and j are both serial numbers, belong to V (set of vehicle nodes) and R (set of fog node resources) respectively, X ij is the data offloading rate of the vehicle node and the fog node resource, and the data rate of the edge vehicle node and the fog node resource is considered in a symmetric mode, and Y ij represents the data offloading rate between fog node resources.

[0110] Data offloading rate X in wireless state is calculated according to Shannon-Hartley theorem ij :

[0111]

[0112] where T p is the transmit power, N0 is the spectral density of the disturbance level, B w is the bandwidth of the channel.

[0113] 2. Vehicle edge computing task model

[0114] In step b, the process of sensing task decision is as follows: for each edge vehicle node v∈V, a computing task is formed by a MIPS (millions of instructions per sec) mode, input data (kbs) and task set S triplets, and each task t can be represented as where s is the input computing size, and c is the number of CPU cycles required for completion.

[0115] 3. Vehicle edge computing energy consumption model

[0116] In step b, the energy consumption model is mainly used to analyze the energy consumption in local computing and task offloading.

[0117] 1) Local computing mode: the resource amount C i used by the i-th task is

[0118]

[0119] where v i is the power consumption rate of each CPU cycle, and the value is 10 -11 (P ci ) 2 ; P ci is the internal computing capacity of the resource that can be provided for the task i∈FN r , c x is the required resource of the subtask, e is the natural logarithm, Q is the task queue length, and x is the serial number of a subtask belonging to Q.

[0120] 2) Offloading computing mode: the resource amount C i of the i-th resource is calculated as

[0121]

[0122] where T p is the transmit power, and v i is the power consumption rate of the CPU cycle, which is 10 -11 (Pci ) 2 ; FN i is the set of fog nodes, e i is the execution cycle required by the task, is the CPU cycle waiting for the task queue to execute.

[0123] 4. Vehicle edge computing latency model

[0124] In step b, according to the computing strategy Od ij , the task is executed locally using the on-board computing unit, and additional resource configuration is provided by the nearby fog node. Therefore, the actual task offloading mode can be divided into executing the task locally or partially delegating to the nearby fog node.

[0125] 1) Local computing mode: the computing time T i is calculated as

[0126]

[0127] where P ci is the computing capacity of the specified resource, e i is the number of CPU cycles required to complete the task,

[0128] is the total number of waiting CPU cycles in the task queue executed locally at the edge vehicle node.

[0129] 2) Offloading computing mode: when the i-th task is delegated to the nearby fog node, the computing time T i is calculated as:

[0130]

[0131] Od ij represents the data offloading rate between the edge node and the fog node, P cj represents the computing energy consumption of allocating resources to the fog node, s i is the task input size, e i is the execution cycle required by the task, is the CPU cycle waiting for the task queue to execute.

[0132] 5. Vehicle edge computing system load model

[0133] In step b, the system load model is used to estimate the demand for computing tasks and the current workload of the device, and according to the load determination result, the two models of local computing and offloading computing are processed.

[0134] 1) Local computing mode: the local overhead is: where and are time and energy coefficients for weighted overhead computation, satisfying and

[0135] 2) Offloading computation mode: the offloading overhead is where and are time and energy coefficients for weighted overhead computation, satisfying and

[0136] 6. Vehicle edge computing task offloading model

[0137] In step c, the process of completing the task offloading decision is as follows:

[0138] After completing the task computation in the space L, it is decided whether to process the task by the fog node or locally, and the task offloading mode is based on the completion time T i The decision-making process can be represented by the following formula:

[0139]

[0140] where 0 represents local execution, 1 represents that the task has timed out when it is offloaded, and ∩ represents.

[0141] The task offloading mode with a prediction mechanism adopted in this embodiment guides the execution of the task offloading decision through the prediction of the position and load of the vehicle, which helps the digital twin assisted mobile edge computing system to perform efficient resource management.

[0142] In step d, when the system optimization goal is to minimize the system overhead, it is particularly important to ensure the workload of the fog node while maintaining the quality of service of the node and avoiding frequent task delivery. The completion of the task offloading decision can be represented as:

[0143]

[0144] where E i represents the efficiency of the i-th fog node, W i represents the corresponding workload, and the workload is limited by the fog node capacity at this time. λ t represents the task arrival rate. Task t is processed by the fog node or the local node in the computation space L with a probability of p, and ∩ represents the task deadline. It is described by , where ∑ l∈L ρ(l) = 1, and the actual processing time of task i needs to meet this task deadline.

[0145] Formula 9 belongs to step d, and its purpose is to determine the "fog node capacity s.t. and the constraint of “task deadline The task offloading decision is executed only if the constraint of formula 9 is satisfied, and the step d can be executed by the fog node.

[0146] The advantages of the scheme provided in the embodiment are compared with various vehicle edge computing methods disclosed in the prior art, and the results are shown in the following table.

[0147] Table 1 Comparison and summary of existing schemes and the scheme of the embodiment

[0148]

[0149] Note: In the above table, the meaning of “Internet of Things communication (resource reachable)” is that each node (such as devices, sensors, servers, etc.) in the Internet of Things network can implement data sending, receiving and processing according to the predetermined protocol and mechanism, and ensure that resources can be accessed and used in time and accurately when needed. Some of the prior art do not consider the problem of resource reachability.

[0150] The literature corresponding to each existing method is as follows:

[0151] Existing method one: Z. Wang, R. Gupta, K. Han, H. Wang, A. Ganlath, N. Ammar, and P. Tiwari, “Mobility digital twin: Concept, architecture, case study, and future challenges,” IEEE Internet of Things Journal, 2022.

[0152] Existing method two: T. Liu, L. Tang, W. Wang, Q. Chen, and X. Zeng,

[0153] “Digital-twin assisted task offloading based on edge collaboration in the digital twin edge network,” IEEE Internet of Things Journal, vol. 9, no. 2, pp. 1427-1444, 2021.

[0154] Existing Method 3: A. A. Alahmadi, T. E. El-Gorashi, and J. M. Elmirghani, “Energy efficient and delay aware vehicular edge cloud,” in 2020 22 nd International Conference on Transparent Optical Networks (ICTON). IEEE, 2020, pp. 1 - 4.

[0155] Existing Method 4: P. Dai, K. Hu, X. Wu, H. Xing, and Z. Yu, “Asynchronous deep reinforcement learning for data-driven task offloading in VEC - empowered vehicular networks.” in INFOCOM, 2021, pp. 1 - 10.

[0156] Existing Method 5: W. Zhan, C. Luo, J. Wang, C. Wang, G. Min, H. Duan, and Q. Zhu, “Deep - reinforcement - learning - based offloading scheduling for vehicular edge computing,” IEEE Internet of Things Journal, vol. 7, no. 6, pp. 5449–5465, 2020.

[0157] Existing Method 6: H. Zhang, Y. Yang, X. Huang, C. Fang, and P. Zhang,

[0158] “Ultra - low latency multi - task offloading in mobile edge computing,” IEEE Access, vol. 9, pp. 32 569–32 581, 2021.

[0159] Existing Method Seven: X. Xu, B. Shen, S. Ding, G. Srivastava, M. Bilal, M. R. Khosravi, V. G. Menon, M. A. Jan, and M. Wang, “Service offloading with deep q-network for digital twinning-empowered internet of vehicles in edge computing,” IEEE Transactions on Industrial Informatics, vol. 18, no. 2, pp. 1414-1423, 2020.

[0160] Existing Method Eight: D. Van Huynh, V.-D. Nguyen, S. R. Khosravirad, V. Sharma, O. A. Dobre, H. Shin, and T. Q. Duong, “Urllc edge networks with joint optimal user association, task offloading and resource allocation: A digital twin approach,” IEEE Transactions on Communications, 2022.

[0161] Existing Method Nine: J. Wang, J. Hu, G. Min, W. Zhan, Q. Ni, and N. Georgalas, “Computation offloading in multi-access edge computing using a deep sequential model based on reinforcement learning,” IEEE Communications Magazine, vol. 57, no. 5, pp. 64-69, 2019.

[0162] Existing method ten: X. Yuan, J. Chen, N. Zhang, J. Ni, F. R. Yu, and V. C. Leung, "Digital twin-driven vehicular task offloading and IRS configuration in the internet of vehicles," IEEE Transactions on Intelligent Transportation Systems, 2022.

[0163] As shown in the above table, unlike other research results and solutions in this field, this solution is the first to handle task offloading demand in a predictive mode under the fog / edge mode, and is the first model to have the three features of digital twin application, minimum latency optimization mode, and resource accessibility.

[0164] In summary, the vehicle edge computing predictive task offloading method and system of the present application combines the technologies of digital twin and fog combination, through the method of predictive task offloading, can reduce the task completion delay, improve the balance of fog node resource allocation, adapt to virtual-real error and improve the calculation rate of VEC system, therefore, the present application has good application prospect.

Claims

1. A method for digital twin assisted vehicle edge computing predictive task offloading, characterized in that, the following steps are performed at the fog node: Step 1: perceiving all task offloading demands in the domain; Step 2: receiving the available computing resource state from the edge vehicle and forwarding it to the digital twin module; Step 3: performing parameter simulation by the digital twin module, determining whether the fog node accepts the task offloading demand by the parameter simulation result and the actual resource inventory, and obtaining the task offloading decision; Step 4: for the accepted task offloading decision, the edge vehicle offloads the task to the fog node for task execution, and then sends the output result back to the edge vehicle; Step 3 specifically comprises the following steps: Step 3.1: inputting the historical task offloading data volume generated in the domain into the integrated artificial intelligence model of the digital twin model; Step 3.2: calculating the computing time required for executing the vehicle task at the next time point in combination with the vehicle movement mode, the available resource state and the historical task offloading data volume; Step 3.3: comparing the computing time of the vehicle task at the next time point with the task waitable delay contained in the task offloading demand, and if the computing time of the vehicle task at the next time point is lower than the task waitable delay, the fog node accepts the task offloading demand, otherwise, it is rejected; The standard for determining whether the fog node accepts the task offloading decision also includes that the edge vehicle is directly connected with the fog node, and the fog node can accept the task offloading decision when executing the task, otherwise, it is rejected.

2. The digital twin assisted vehicle edge computing predictive task offloading method of claim 1, wherein: In Step 1, the fog node is responsible for maintaining a list of edge vehicles accessing and leaving within the communicable network range of the fog node; the edge vehicle is initialized when accessing, and sends its task offloading demand to the fog node.

3. The digital twin assisted vehicle edge computing predictive task offloading method of claim 1, wherein: In Step 3, the digital twin module is configured to create the network topology of all connected nodes and the task interaction relationship, perform parameter simulation on the task offloading decision and data transmission mode of the actual fog node or edge node using the digital twin model, and realize prediction of the task computing capability of the fog node.

4. The digital twin assisted vehicle edge computing predictive task offloading method of claim 1, wherein: In Step 3.1, the artificial intelligence model algorithm is selected from LSTM, RNN, CNN, Bayesian method or ARIMA time series model.

5. The digital twin assisted vehicle edge computing predictive task offloading method of claim 1, wherein: Step 4 specifically comprises the following steps: Step 4.1: when the task offloading decision is accepted, the edge vehicle offloads the task to the fog node, and all accepted tasks are saved in the local queue of the fog node until the resource meets the demand for stable execution of the task; Step 4.2: after the task execution, all output results are stored in the result queue of the fog node; Step 4.3: the scheduling program of the fog node sends the output result back to the edge vehicle. 6.The method according to claim 1, characterized in that: the following steps are performed locally at the edge vehicle: Step a: the edge vehicle generates a task and stores the task in the local queue; Step b: The resource manager of the edge vehicle perceives the task offloading decision basis in real time, calculates the demand of the task and the current workload of the device, decides whether the task is processed locally or by the fog node, and obtains the task offloading demand for the task processed by the fog node; the task offloading decision basis includes vehicle metadata, task data, fog resources and task deadline information; Step c: temporarily store the task offloading demand in the offloading queue; Step d: send the task offloading demand when the edge vehicle is connected with the fog node.

7. A system for implementing the method of claim 1-6 for digital twin assisted vehicle edge computing predictive task offloading, characterized in that, Comprise: a fog node and at least one edge vehicle; The fog node is connected with a digital twin module.

8. A computer-readable storage medium, characterized in that, A computer program is stored thereon, which is used to implement the digital twin assisted vehicle edge computing predictive task offloading method according to any one of claims 1-6.

Images