Dynamic resource slicing method for air-ground integrated vehicle network
By designing a multi-time scale dynamic resource slicing framework and load balancing algorithm in an integrated air-ground vehicle network environment, the problem that the static slicing method cannot adapt to the dynamic network environment is solved, the flexibility and efficient utilization of resource management are achieved, and task completion is improved rate and resource utilization rate.
Patent Information
- Application Number
- CN202210619340.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-02
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-06-02
AI Technical Summary
In the integrated air-ground networking environment, the static network slicing method cannot adapt to the time-varying network environment. Frequent redistribution of slice resources increases network control overhead, and drones carry resources on demand, resulting in a decrease in the duration of connection between vehicles and base stations. Increased resource management difficulty.
Design a multi-time scale dynamic resource slicing framework to collaborate resource slicing at large time scales with workflow scheduling at small time scales. Through the task offloading process based on queuing theory modeling, avoid overloading of a single base station, and propose a load balancing algorithm , optimize workflow scheduling to adapt to the dynamics of network service types and road network traffic.
It improves resource management flexibility, reduces signaling overhead, ensures fairness of network services and efficient utilization of global resources, and significantly improves the number of task completions and system resource utilization.
Smart Images

Figure CN115103376B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of vehicle wireless communication, and specifically is a dynamic resource slicing method for air-ground integrated vehicle networking. Background Art
[0002] The Internet of Vehicles (IoV) is based on vehicle wireless communication technology, connecting vehicles, roadside units (ROUs), base stations, and service providers into an organic whole to achieve real-time information sharing[1]. Vehicle users can obtain services such as autonomous driving, path planning, collision warning, in-vehicle entertainment, and high-definition map downloads[2]. The IoV is a dynamic environment. Relying solely on the ground radio access network (RAN) is difficult to provide vehicle users with a full range of services anytime and anywhere, and there are problems such as low resource utilization and difficulty in providing stable quality of service (QoS)[3]. The air-ground integrated network architecture is considered to be an effective solution to improve the management flexibility, resource utilization, and network coverage of the IoV[4]. In this network, low-power drone small base stations are deployed on demand in the road network, and the deployment location can be dynamically adjusted. Drone small base stations have the advantage of light-of-sight communication, which helps to enhance network coverage, improve link quality, and resource utilization. In addition to providing communication services, drone small base stations can also serve as aerial mobile edge computing (MEC) servers to provide services such as content caching and computing offloading for vehicles. [5] .
[0003] Network slicing[6] is an important technological innovation in network architecture and service models in the 5G era. By dividing the physical network into multiple logically independent virtual networks (i.e., slices), multiple operators can share the resources of the same physical network, thereby improving network management flexibility and reducing infrastructure expenditures and operating costs. Network Function Virtualization (NFV)[7] and Software-Defined Networking (SDN)[8] are the supporting technologies for network slicing. On the RAN side, base station functions include radio access and processing, which are used to create wireless connections and allocate resources. In wireless NFV, functions such as radio access run on the base station in the form of software instances and are managed by a centralized controller. By collecting terminal request information, the controller creates slices according to QoS requirements and schedules network slice resources based on real-time network traffic or topology information.
[0004] The Internet of Vehicles has a natural reliance on network slicing technology. This is because multiple services coexist in the Internet of Vehicles and many applications have strict latency and reliability requirements. However, the design of resource slicing methods under the air-ground integrated Internet of Vehicles faces many challenges. First, the static network slicing method cannot be applied to time-varying network environments, but frequent slice resource reallocation will also increase network control overhead. Secondly, drones carry resources and deploy them on demand, while network slicing requires resources to be reserved for drones. Finally, the Internet of Vehicles is characterized by high-speed mobility. The connection duration between the vehicle and the base station is therefore reduced, increasing the difficulty of workflow scheduling and network slicing resource management.
[0005] In the existing technology, RAN slicing has significant advantages in reducing network operating costs and improving resource utilization. Reference [9] proposed a latency-aware 5G network slicing node and link mapping cost minimization algorithm. A two-level queue dynamic scheduling model is adopted to achieve balanced control of mapping cost and system latency. Based on specific service traffic statistics, reference
[10] proposed a heterogeneous wireless network downlink spectrum resource slicing framework to provide differentiated QoS guarantees for machine type devices and mobile user devices. The scheme uses an iterative optimization method to solve the joint decision-making problem of resource allocation and device access selection, aiming to maximize network utility. On this basis, reference
[11] incorporates a transmit power adaptive adjustment mechanism and designs a spectrum resource slicing strategy based on multi-access edge computing to provide customized services for vehicle network applications. In general, the model-based optimization method is suitable for small-scale RAN slicing and cannot adapt to the dynamically changing air-ground integrated wireless network environment.
[0006] How to allocate RAN resources to support task offloading will also affect the overall resource utilization. Autonomous driving tasks often have differentiated QoS characteristics. If the allocation of radio resources cannot meet the requirements of task transmission rate, latency or reliability, the load balancing of computing may not be achieved. Reference
[12] studies the joint communication and computing time allocation problem of vehicle network collaborative computing offloading, and jointly optimizes the task offloading resources, local task execution resources and vehicle auxiliary task migration resources to achieve the overall maximum reliability of task computing. Reference
[13] designs an adaptive computing offloading method suitable for edge computing to optimize the task offloading delay and resource utilization of edge computing systems in order to select the offloading destination of computing tasks. Reference
[14] In order to narrow the gap between the service quality requirements of applications and the limited computing resources of base stations and improve the resource utilization of devices and base stations, an online optimization problem combining task offloading, device-base station association and base station sleep scheduling is proposed to minimize the overall energy consumption of devices and base stations.
[0007] Different from the traditional ground network resource management, resource slicing for air-ground integrated vehicle networking needs to consider the coordinated management of drone small base stations, vehicle-base station association, and workflow scheduling. This drives the present invention to study a joint method of vehicle association mode and dynamic resource slicing to cope with the dynamically changing network environment and provide differentiated QoS guarantees for vehicle applications. Summary of the invention
[0008] With the advantages of mobile deployment and line-of-sight communication, drone small base stations have the potential to enhance and expand Internet of Vehicles services. This paper proposes a dynamic resource slicing method for air-ground integrated Internet of Vehicles, aiming to maximize the Internet of Vehicles task completion rate on the basis of providing differentiated quality-of-service (QoS) guarantees.
[0009] The present invention provides a dynamic resource slicing method for an air-ground integrated vehicle network, and its application network scenario includes multiple base stations, multiple vehicles, and a controller; the base stations are divided into ground base stations and drone base stations; the base stations receive all task requests from vehicles within the coverage area. According to the two types of tasks, delay-sensitive and delay-tolerant, the spectrum resources of each base station are divided into two spectrum resource slices for serving the two types of tasks respectively; the slicing time window T is divided into multiple scheduling windows t of equal size; within the slicing time window T, the base station allocates spectrum resources according to the slicing scheme; within the scheduling window t, the controller performs workflow scheduling on the received tasks;
[0010] The controller formulates a spectrum resource segmentation strategy based on the service type and resource requirements of the tasks in the road network, and slices the spectrum resources according to the segmentation strategy;
[0011] The steps of workflow scheduling are as follows: the vehicle submits a task request to the base station; the base station receives the task and unloads it to the task unloading queue in the controller; the base station processes the task according to the workflow decision of the controller; the base station transmits the task processing result back to the vehicle, and the base station uploads the task data to the controller;
[0012] The spectrum resource segmentation strategy is formulated based on the scheduling data of the historical window, which reduces the calculation pressure of the base station; on the other hand, it is based on the actual situation of the road network to cope with the dynamic changes of traffic;
[0013] Workflow scheduling needs to consider multiple factors, such as task service type, QoS restrictions, the number of allocatable resources in the base station, and queuing information;
[0014] By dynamically dividing the spectrum resources of each base station and optimizing workflow scheduling, we can adapt to the differentiation of service types in the network and the dynamic nature of road network traffic.
[0015] The spectrum resource segmentation strategy changes dynamically according to actual conditions;
[0016] The dynamic nature of the network can be adapted by dynamically dividing the spectrum resources of each base station and optimizing workflow scheduling.
[0017] The technical contributions of the present invention mainly include:
[0018] Firstly, a multi-timescale dynamic resource slicing framework is designed to organically coordinate resource slicing on a large time scale with workflow scheduling on a small time scale, improve resource management flexibility, and reduce unnecessary signaling overhead.
[0019] Secondly, the task offloading process is modeled based on queuing theory to avoid overloading of a single base station and ensure the fairness of network services.
[0020] Finally, a load balancing algorithm is proposed to determine the workflow scheduling, which ensures QoS while taking into account the efficient utilization of global resources.
[0021] Simulation results show that compared with existing schemes, the proposed scheme can improve the number of completed tasks, system resource utilization, and network service fairness. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 : Schematic diagram of air-ground integrated vehicle network;
[0023] Figure 2 : Schematic diagram of a multi-timescale service-oriented resource slicing framework;
[0024] Figure 3 : Schematic diagram of workflow scheduling framework;
[0025] Figure 4 : Schematic diagram of the effect of the number of available resource blocks on the task completion rate;
[0026] Figure 5 : Schematic diagram of the impact of vehicle density on global resource utilization;
[0027] Figure 6 : Schematic diagram of the impact of the proportion of delay-sensitive services on task completion rate;
[0028] Figure 7 : Schematic diagram of the impact of the number of drones on the mission completion rate. DETAILED DESCRIPTION
[0029] 1. Overview of the Invention
[0030] This paper proposes a dynamic resource slicing method for air-ground integrated vehicle networking, the purpose of which is to maximize the task completion rate on the basis of meeting the vehicle application task offloading delay requirements. The main work includes:
[0031] 1) In order to improve the flexibility of RAN resource management, a multi-time scale dynamic resource slicing framework is designed. Resource slicing is performed on a large time scale, and workflow scheduling is performed on a small time scale to enhance the stability of slice resource provision and reduce signaling overhead.
[0032] 2) Based on queuing theory, the task offloading process is characterized and the task processing process of each base station is modeled as an M / M / 1 queue model. By monitoring the task queuing situation, the overload of a single base station is avoided and the fairness of network services is achieved.
[0033] 3) Considering the high-speed mobility of vehicles, differentiated QoS requirements, task queuing and offloading delay, a load balancing algorithm is developed to determine the workflow scheduling in real time according to the network environment, while ensuring the requested QoS and taking into account the efficient utilization of global resources.
[0034] The simulation results show that the proposed method can effectively adapt to the spatiotemporal changes of IoV traffic and provide stable services for vehicle users. Compared with existing solutions, it can significantly improve resource utilization and the probability of completing task results.
[0035] The following sections are arranged as follows:
[0036] Section 2 describes the proposed system model in detail;
[0037] Section 3 constructs spectrum resource slicing and workflow scheduling as a constrained stochastic optimization problem;
[0038] Section 4 decouples the stochastic optimization problem into resource slicing subproblems and workflow scheduling subproblems, and proposes a load balancing algorithm;
[0039] Section 5 introduces the experimental parameter settings and simulation results;
[0040] 2 System Model
[0041] 2.1 Network Scenarios and Assumptions
[0042] like Figure 1 As shown in FIG, the present invention considers an air-ground integrated vehicle networking scenario, which includes B ground base stations, K drone small base stations and I vehicles. The sets of ground base stations, drone base stations and vehicles are represented as In order to reduce the latency of vehicles obtaining services and improve service efficiency, a controller assisted by MEC is added. The controller and the base station / drone are connected wirelessly, and all requests within the coverage of the base station / drone can be handled by the controller. The requests randomly generated by the vehicle are unloaded to the controller by the base station. The controller assigns tasks in real time according to the network environment and hands them over to the base station for processing. Finally, the base station transmits the processing results back to the vehicle. The task arrival rate of the vehicle is denoted as λ. The set of tasks is denoted as
[0043] The vehicle needs to be within the coverage of the base station to connect to the base station to offload tasks. The effective coverage radius of base station j is defined as δ j ,in The three-dimensional coordinates of base station j are expressed as (x j ,y j ,z j ), the three-dimensional coordinates of vehicle i are expressed as (x i ,y i ,z i ). Define the set of vehicles within the coverage of base station j as Expressed as
[0044]
[0045] Then the coverage of vehicle i within base station j can be expressed as The vehicle is equipped with two signal receivers, one for connecting to the ground base station and the other for connecting to the drone. However, each vehicle can only be associated with a single base station (ground base station or drone base station) to offload tasks. If the vehicle is within the coverage of multiple base stations at the same time, the base station connection is selected according to the workflow scheduling decision.
[0046] 2.2 Dynamic Resource Slicing Framework
[0047] Vehicle aggregation and high-speed driving will cause dynamic changes in the association pattern, which in turn triggers network resource scheduling. Considering the above problems, the present invention designs a long-term and short-term collaborative optimization RAN slicing framework to jointly optimize resource partitioning and vehicle association patterns. Figure 2 As shown in the figure, the present invention considers two typical types of IoV requests, namely, delay-sensitive requests and delay-tolerant requests. The former corresponds to autonomous vehicle platoon control
[15] , and its delay constraint is only 100-150ms; the latter's typical application includes autonomous vehicle high-definition map downloading
[16] , and its delay requirement is relatively loose.
[0048] Request type o=1 (o=2) corresponds to delay-sensitive (delay-tolerant) tasks. The physical resources (spectrum resources) of each base station are divided into two service-oriented resource slices, namely slice 1 and slice 2, which support delay-sensitive tasks and delay-tolerant tasks respectively. Each slice contains a task processing queue. Define w j is the total number of channels held by base station j. The controller uses the same ratio {α w ,β w} Divide the spectrum resources of each base station into two parts, namely α w +β w = 1. Define w j,1 and w j,2are the number of subchannels of the delay-sensitive and delay-tolerant slices allocated to base station j, respectively.
[0049] Due to the temporal and spatial variations of traffic flow, the spectrum resource segmentation strategy needs to be customized according to the actual traffic conditions of the road network. Figure 2 As shown, the set of slice time windows is defined as In each slicing window (i.e., slicing time window), the controller formulates a slicing scheme based on the collected historical request information. Each base station allocates spectrum resources according to the slicing scheme. Then, on a small scale, the present invention divides each slicing window into multiple scheduling windows of equal size, i.e., T = {1, ..., t, ...}. In the scheduling window t, the controller schedules the received tasks; each base station processes the tasks according to the workflow decision; the base station transmits the processing results of the tasks back to the vehicle; the base station uploads the task data to the controller.
[0050] 2.3 Communication model
[0051] The base station allocates resources in the same slice to the associated vehicles in an orthogonal manner according to the request type. The transmission power of vehicle i is represented as P i The transmission power of base station j is denoted as P j . Define σ 2 is the average background noise. The spectrum resources allocated by the base station to the vehicle are orthogonal to each other and will not interfere with each other. The interference to the vehicle only comes from the transmission signals of other base stations. If the number of subchannels allocated to vehicle i for request type o is w i,o , then the uplink transmission rate of the request type o task generated by vehicle i within the coverage area of base station j is quantified as
[0052]
[0053] Wherein, j′ represents the remaining base stations in the base station set after removing j. i,j represents the channel gain between vehicle i and base station j, which is calculated according to reference
[17] .
[0054] When the vehicle receives the feedback result from the base station, it is also only subject to interference from other base stations. Therefore, the downlink transmission rate from base station j to vehicle i for the task of type o is quantified as
[0055]
[0056] 2.4 Workflow Scheduling Model
[0057] The present invention uses queuing theory
[18] to characterize the process of base station processing tasks, and obtains the QoS constraints of delay-sensitive and delay-tolerant tasks through analysis. Within the slice window T, the task m generated by vehicle i contains the task data size (bits) and the delay requirement for task processing completion, which are expressed as The derivations in this section are all performed within a slice window T. For ease of reading, T is omitted in the subsequent parameters.
[0058] Workflow scheduling framework such as Figure 3 As shown in the figure, both the controller and the base station have two queues: delay-sensitive and delay-tolerant. The queue management modes are:
[0059] 1) The tasks generated by the vehicle are received by each base station and unloaded to the task unloading queue in the controller;
[0060] 2) Following the principle of first come first served, the controller delivers the task to the base station for processing;
[0061] 3) After receiving the task, the base station puts it into the processing queue and allocates physical resources for processing;
[0062] 4) The base station transfers the processed results to the vehicle that issued the request.
[0063] 2.4.1 Task offloading delay modeling
[0064] The task offloading delay represents the time it takes for a task to be transmitted from a vehicle to a base station j. The controller and the base station are connected via a wired network. Within the area covered by the controller, the average time it takes for a task with a request type o to be transmitted from a vehicle to a base station is quantified as
[0065]
[0066] The arrival of a single vehicle's mission is modeled as a Poisson process, and the arrival of missions received by the base station is also modeled as a Poisson process. Define the binary variable a i,j =1 represents that vehicle i is associated with base station j. In other words, the arrival rate of request type o tasks can be expressed as
[0067]
[0068] where λ i,o It represents the arrival rate of vehicle i generating tasks of request type o.
[0069] The unloading queue only processes one task at a time. The task unloading process is modeled as an M / M / 1 queue model. The entry of the unloading queue is determined by the task arrival rate, and the exit of the unloading queue is determined by the base station transmission. When the queue entry rate is greater than the exit rate, the tasks in the queue will continue to accumulate and cause the queue to overflow. The service intensity of the queue reflects the busyness. The service intensity of the unloading queue with request type o in base station j is defined as
[0070]
[0071] Where μ o The average time it takes for a task with request type o to be transmitted from the vehicle to the base station
[0072] In order to maintain the stability of the unloading queue (to prevent the queue from overflowing), formula (6) needs to satisfy
[0073]
[0074] The position of the task in the queue is calculated from the head of the queue. The set of the first n tasks in the offload queue of base station j is represented as Ω j (n). Assuming that task m with request type o generated by vehicle i is unloaded to base station j and its rank in the unloading queue is n, the unloading delay of the task is calculated as
[0075]
[0076] 2.4.2 Task Handover Delay Modeling
[0077] like Figure 3 As shown, after each task is calculated in the processing queue of the base station, the result is directly transmitted back to the vehicle by the base station. Based on formula (3), the handover delay of task m of request type o in base station j back to vehicle i is expressed as
[0078]
[0079] The task delay consists of unloading delay and handover delay. From (8) and (9), the task delay of request type o task m generated by vehicle i is
[0080]
[0081] The completion of a delay-sensitive task requires two conditions to be met. The delay requirement must be met; and the task must be completed before the vehicle leaves the coverage of the base station. i,j,m The distance from the edge of base station j when vehicle i generates task m. Then the maximum travel time within the coverage of base station j when vehicle i generates task m can be calculated as
[0082]
[0083] Among them, v i represents the driving speed of vehicle i. The constraint on the completion of delay-sensitive tasks can be expressed as
[0084]
[0085] 3 Problem Modeling
[0086] By dynamically slicing the spectrum resources of each base station and optimizing the workflow scheduling to adapt to the dynamic nature of the network, the goal of the present invention is to maximize the number of tasks completed on the basis of meeting differentiated QoS requirements. The performance optimization within the slicing window T depends on the spectrum resource slicing strategy and the workflow scheduling strategy, which are respectively defined as the spectrum resource slicing strategy
[0087] and workflow scheduling strategies for Define the following binary variables
[0088]
[0089] Indicates whether the task processing meets the delay constraint after vehicle i connects to base station j in the tth scheduling window. If the delay requirement can be met, then otherwise
[0090] When the task is completed under the condition of satisfying the delay constraint, the system obtains the corresponding benefit. Correspondingly, if the task is not completed, the system incurs the corresponding loss.
[0091] Definition 1: In the tth scheduling window, when the task is completed and the delay requirement is met, the total reward U obtained by the system is (t)
[0092]
[0093] where u j,o ∈(0,1) represents the corresponding profit factor of the task with request type o on base station j.
[0094] Definition 2: When a task fails to meet the delay requirement in the tth scheduling window, the total loss H generated by the system is ( t)
[0095]
[0096] where h j,o ∈(0,1) represents the corresponding loss factor of the task with request type o on base station j.
[0097] There are some differences in the time and space of vehicle task release on the road. Traffic flow is divided into peak period and idle period, so the number of tasks on the same road section will change over time. Due to traffic control by traffic lights, vehicles often gather at intersections and do not stay too long in the middle of the road.
[0098] Under the premise of meeting QoS requirements, the goal of this invention is to enable the system to complete more vehicle tasks in the long term. With the goal of maximizing the number of vehicle tasks completed, the dynamic RAN slicing problem is modeled as
[0099]
[0100]
[0101]
[0102]
[0103] The essence of problem P0 is to coordinate the allocation of spectrum resources of each base station and the workload of the region in an online manner so that the average number of tasks completed by the system in the slice window T is maximized. Among them, constraint (15a) ensures that the number of sub-channels allocated to each base station j is a positive number. Constraint (15b) ensures that the number of channels allocated to vehicles by each base station does not exceed the total number of channels it holds. Constraint (15c) ensures that each vehicle can only connect to a unique ground base station / UAV small base station, and cannot connect to multiple at the same time. Constraint (15d) ensures the stability of the queue in the queuing system. At the same time, it also shows that the spectrum resource slicing decision and the workflow allocation decision are coupled, that is, the coupling constraint.
[0104] 4 Problem decoupling and algorithm design
[0105] For ease of processing, P0 is decomposed into spectrum resource slicing subproblems and workflow scheduling subproblems. The spectrum resource slicing subproblem solves the optimal slicing decision by giving the workflow scheduling decision.
[0106] Essentially, it is a constrained extremum problem, which is suitable for solving using traditional optimization methods. The workflow scheduling subproblem is to find the optimal workflow scheduling decision given the slicing decision. This subsection designs a load balancing algorithm to solve the scheduling problem.
[0107] 4.1 Spectrum Resource Slicing Subproblem on Large Time Scale
[0108] The spectrum resource slicing sub-problem is to make scheduling decisions for a given workflow. Under the condition of optimizing resource slicing decision To maximize the number of requests completed by the system.
[0109]
[0110] st constraints (15a), (15b) and (15c)
[0111] According to the definition of system benefits in formulas (13) and (14), given resource slices, resource allocation determines the number of tasks that the base station can process. The condition for task completion is to allocate sufficient resources to meet the delay requirements, and each task is independent of each other. The essence of the spectrum resource slicing subproblem is to maximize the number of completed tasks in each slice window. The decisions of each slice window are independent and do not affect each other. In reality, the traffic flow will not experience continuous large fluctuations, and the traffic flow in adjacent slice windows is similar. In this regard, the controller can use the actual workflow decision of the previous slice window to optimize the resource slice. The controller records the workflow decisions in each slice window to provide a reference for optimizing the resource slice decision for the next slice window, so as to maximize the number of completed requests in each slice window. The above problem corresponds to the following one-time optimization problem:
[0112]
[0113]
[0114] Constraints (15b) (16c)
[0115] Problem P2 is to solve the extremum problem of a multivariate function subject to multiple conditions, which can be solved using the Lagrange multiplier method. This method transforms an optimization problem with multiple variables and multiple constraints into an extremum problem of an unconstrained set of equations with multiple variables. Problem P2 is transformed into P3.
[0116]
[0117] Given a workflow decision, the controller can calculate the specific number of tasks that each base station handles. Then, the spectrum resource slicing problem is constructed based on the spectrum resource requirements and latency requirements of the tasks and the spectrum resource holdings of each base station and drone. Calculating P3 can obtain an optimal spectrum resource slicing solution, under which the efficiency of workflow scheduling will be improved and the system resource utilization will also be improved.
[0118] 4.2 Small-time-scale workflow scheduling subproblem
[0119] The goal of the workflow scheduling subproblem is to maximize the number of completed requests under the constraint of queue stability, that is, given the spectrum resource slicing decision Under the condition of To maximize the number of tasks completed by the system.
[0120]
[0121] st constraint 15(d)
[0122] In problem P1, the resource allocation of each slice window is independent of each other. Correspondingly, the scheduling of workflows under fixed resource slice decisions in each slice window is also independent of each other. Therefore, when solving problem P4, the long-term optimization problem can be decomposed into short-term optimization problems in each scheduling window.
[0123] Workflow scheduling needs to consider the number of tasks waiting to be processed in the system, the spectrum resource requirements, waiting time and delay requirements of each task, as well as the queue processing information of each base station and drone. In order to improve resource utilization and the number of completed requests, this subsection designs a load balancing algorithm to solve the workflow scheduling problem, focusing on the efficient utilization of the entire network resources.
[0124] The controller not only needs the information of each task in the offload queue, but also needs the spectrum resource scheduling of each base station and drone and the task information in the internal processing queue. The controller allocates the tasks in the offload queue according to the first-come-first-served strategy. The specific workflow scheduling is expressed as Algorithm 1.
[0125]
[0126] In Algorithm 1, line 6 determines whether the processing time of the task in each processing queue meets the delay requirement; line 10 determines whether the remaining spectrum resources of each base station can meet the current task requirements; line 14 determines whether the maximum driving time of the current task in each base station meets the delay requirement. In the initial stage, the connection option for the task is all base stations. The algorithm in this section comprehensively considers multiple factors such as the spectrum resource usage of each base station, the task processing time, and the connection time between the vehicle and the base station, and makes workflow scheduling decisions after multiple screenings. It not only improves the utilization of system resources, but also ensures the fairness of task processing.
[0127] 4.3 Joint Optimization Framework
[0128] This subsection proposes a joint optimization framework to integrate the spectrum resource slicing subproblem with the workflow scheduling subproblem. First, the system divides the length of the slicing window according to the task information in the historical data. After the slicing window is determined, the workflow scheduling decision within the T-1th slicing window is As a known condition for solving problem P3, and solving the resource slicing decision Workflow scheduling decision for the first slice window Given by historical data. Divide the slice window T into multiple scheduling windows of equal size T = {1,...,t,...}. In each scheduling window, make resource slice decisions As a known condition for solving problem P4, the workflow scheduling decision in each scheduling window is obtained. Each base station processes tasks according to the workflow scheduling decision. At the end of the last scheduling window, the system integrates the workflow scheduling decision in each scheduling window into the workflow scheduling decision of slice window T. And recorded as historical data for use in the T+1th slice window. The joint optimization framework can be expressed as Algorithm 2.
[0129]
[0130] The joint optimization framework enables the long-term operation of resource slicing and workflow scheduling. By using the similarity of traffic flow in adjacent time periods and taking the workflow scheduling decision of the previous slicing window as a known condition, the spectrum resource slicing decision is obtained. This not only reduces the system's computing tasks, but also improves the applicability of slicing decisions.
[0131] 5 Experimental design and results analysis
[0132] The effectiveness of the proposed scheme is verified by simulation method. Consider a 1000-meter-long four-lane highway, which contains 2 macro base stations and 3 drone small base stations. In order to make the simulation close to the real environment, the vehicle flow data source selected by this invention is the OpenITS open data platform①. The vehicle density on the road is set to 0.4 (vehicles / m 2 ). The origin of the coordinates is the starting point of the road. The two macro base stations cover half of the road respectively, with a coverage range of 500m. The drone is pre-set on the road, with a coverage range of 80m. The two macro base stations have the same transmission power of 40dBm, and the three drones have the same transmission power of 0.1W. The delay-sensitive task is the formation control of unmanned vehicles, and the delay-tolerant task is the download of high-definition maps for autonomous vehicles. Other parameters are shown in Table 1.
[0133]
[0134] In order to objectively evaluate the performance, the present invention selects three representative resource management and terminal association strategies for comparison, including:
[0135] Resource management method based on maximum signal-to-noise ratio (Max-SINR)
[19] : The resource slice ratio is evenly divided, and when the controller schedules the workflow, it selects the base station with the maximum signal-to-noise ratio connected to the vehicle.
[0136] Random resource scheduling method (Random): The resource slice ratio is randomly allocated, and the controller randomly selects base stations when scheduling workflows.
[0137] · Drone-first vehicle association method (RSE-online)
[20] : The resource slice ratio is evenly divided, and the controller gives priority to idle drones when scheduling workflows.
[0138] First, the impact of increasing spectrum resources on the task completion rate is evaluated. The simulation environment includes 2 macro base stations and 3 drones. Figure 4 It can be seen that as the number of resource blocks held by the system increases, the system's task completion rate also increases. When the number of available resource blocks is less than 35, the task completion rate of this scheme is better than the other three schemes. When the number of available resource blocks increases to 35 or more, the task completion rates of the other three schemes except random scheduling are almost 100%. This is because when the schedulable resources in the network are insufficient, the performance of each method is different; when the resources in the network are sufficient, all tasks can be satisfied.
[0139] Figure 5 The impact of vehicle density on global resource utilization was evaluated. 2 When the vehicle density increases to 0.3 vehicles / m, the global resource utilization of the four schemes is less than 50%. This is because the tasks in the network are sparse, some base stations are idle, and the resources in the system cannot be fully utilized. In addition, it can be seen that as the vehicle density increases, the global resource utilization continues to increase. Compared with the baseline scheme
[19]
[20] , the resource utilization of this scheme increases by 29% and 10% respectively. When the vehicle density increases to 0.3 vehicles / m 2 After that, the resource utilization of the baseline scheme
[20] and this scheme is significantly higher than that of other schemes. This is because the increase in vehicle density leads to an increase in the number of tasks, and the flexible scheduling of drones effectively increases the system resource utilization. However, the resource utilization cannot be increased to 100%. This is because the vehicle must be within the coverage of the base station to connect to it and unload tasks, and the base station far away from the vehicle cannot provide services for it.
[0140] Figure 6 The impact of increasing the proportion of delay-sensitive services on the task completion rate was evaluated. As the proportion of delay-sensitive services increased, the task completion rate continued to decrease. This is because the QoS restrictions of delay-sensitive services require more resources for task processing. Increasing the proportion of delay-sensitive services is a stress test on the processing capacity of the system. As can be seen from the figure, the task completion rate of this solution is higher than that of other solutions. This is because this solution pays more attention to the scheduling of drones and the utilization of global resources than other solutions. It shows better robustness when facing extreme conditions.
[0141] Figure 7 The vehicle density was evaluated at 0.9 (vehicles / m 2 ) scenario, the impact of the number of drones on the task completion rate. As can be seen from the figure, when the system faces a large number of services, the increase in the number of drones scheduled can effectively increase the task completion rate. Improve the performance of the system. Among them, the random scheduling scheme is not sensitive to the increase of drones, and the task completion rate is less than half. The task completion rates of the benchmark scheme
[19]
[20] and the scheme of the present invention are improved due to the increase of drones. This is because the location of the base station is fixed and cannot handle all tasks in the Internet of Vehicles environment. Tasks at the edge of the base station coverage are often not processed because they are far away from the base station. Drones can solve this problem well. When hot spots appear in the Internet of Vehicles, drones can be dispatched to the hot spots to reduce the pressure on the base stations and improve the service quality of users.
[0142] 6 Conclusion
[0143] The present invention divides time into two scales, large and small, to adapt to the spatiotemporal changes of vehicle network traffic, and improves the robustness of the system processing capability. In view of the two typical requests of vehicles, the spectrum resources are divided into two slices, which realizes the isolation of service QoS and improves the task completion rate of vehicles. Compared with the existing solutions, the load balancing solution proposed by the present invention effectively increases the number of completed tasks, improves the utilization rate of system resources, and realizes the fairness of network services.
[0144] References:
[0145] [1] Zhuang, W., Ye, Q., Lyu, F., Cheng, N., & Ren, J. SDN / NFV-empowered futureIoV with enhanced communication, computing, and caching. Proceedings of the IEEE, 2020, 108(2): 274-291.
[0146] [2]Ning, Z., Hu,
[0147] [3]Zhang,W.,Zhang,Z.,&Chao,H.C.Cooperative fog computing for dealingwith big data in the internet ofvehicles:Architecture and hierarchicalresource management.IEEE Communications Magazine,2017,55(12):60-67.
[0148] [4]Liu,J.,Shi,Y.,Fadlullah,Z.M.,&Kato,N.Space-air-ground integratednetwork:A survey.IEEECommunications Surveys&Tutorials,2018,20(4):2714-2741.
[0149] [5]Cheng,N.,Xu,W.,Shi,W.,Zhou,Y.,Lu,N.,Zhou,H.,&Shen,X.Air-groundintegrated mobile edgenetworks:Architecture,challenges,and opportunities.IEEECommunications Magazine,2018,56(8):26-32.
[0150] [6]Sexton,C.,Marchetti,N.,&DaSilva,L.A.Customization and trade-offsin 5G RAN slicing.IEEECommunications Magazine,2019,57(4):116-122.
[0151] [7]Yi,B.,Wang,X.,Li,K.,&Huang,M.A comprehensive survey of networkfunction virtualization.ComputerNetworks,2018,133:212-262.
[0152] [8]Haque,I.T.,&Abu-Ghazaleh,N.Wireless software defined networking:Asurvey and taxonomy.IEEECommunications Surveys&Tutorials,2016,18(4):2713-2737.
[0153]
[10] Ye Q,Zhuang W,Zhang S,Jin AL,Shen X,Li X.Dynamic radio resourceslicing for a two-tier heterogeneouswireless network.IEEE Transactions onVehicular Technology,2018,67(10):9896-9910.
[0154]
[11] Peng,H.,Ye,Q.,&Shen,X.Spectrum management for multi-access edgecomputing in autonomous vehicularnetworks.IEEE Transactions on IntelligentTransportation Systems,2019,21(7):3001-3012.
[0155]
[12] Omar,H.A.,Zhuang,W.,Abdrabou,A.,&Li,L.Performance evaluation ofVeMAC supporting safetyapplications in vehicular networks.IEEE Transactionson Emerging Topics in Computing,2013,1(1):69-83.
[13] Xu,X.,Zhang,X.,Liu,X.,Jiang,J.,Qi,L.,&Bhuiyan,M.Z.A.Adaptive computation offloading with edgefor5G-envisioned internet of connected vehicles.IEEE Transactions on IntelligentTransportation Systems,2020,22(8):5213-5222.
[0156]
[15] Tsugawa,S.,Jeschke,S.,&Shladover,S.E.A review of truck platooningprojects for energy savings.IEEETransactions on Intelligent Vehicles,2016,1(1):68-77.
[0157]
[16] Javanmardi,E.,Gu,Y.,Javanmardi,M.,&Kamijo,S.Autonomous vehicleself-localization based on abstractmap and multi-channel LiDAR in urbanarea.IATSS research,2019,43(1):1-13.
[0158]
[17] Erceg, V., Greenstein, LJ, Tjandra, SY, Parkoff, SR, Gupta, A., Kulic, B.,... & Bianchi, R. An empirically based path loss model for wireless channels in suburban environments. IEEE Journal on selected areas in communications, 1999, 17(7): 1205-1211.
[0159]
[18] Xue, J., Wang, Z., Zhang, Y., & Wang, L. Task allocation optimization scheme based on queuing theory for mobile edge computing in 5G heterogeneous networks. Mobile Information Systems, 2019, 62(2): 1-3.
[0160]
[19] Ye, Q., Zhuang, W., Zhang, S., Jin, AL, Shen, X., & Li,
[0161]
[20] Tang, L., Zhao, G., Wang, C., Zhao, P., & Chen, Q. Queue-aware reliable embedding algorithm for 5G networkslicing. Computer networks, 2018, 146(9): 138-150.
[0162] Attached Chinese references:
[0163] [9] Tang Lun, Yang Heng, Zhao Guofan, Wang Yaowei, Chen Qianbin. Latency-aware 5G network slice node and link mapping algorithm. Journal of Beijing University of Posts and Telecommunications, 2018, 41(6): 71-77.
[0164]
[14] Yu Bowen, Pu Lingjun, Xie Yuting. Research on task offloading and base station-related collaborative decision-making in mobile edge computing[J]. Journal of Computer Research and Development, 2018, 55(3): 537-550.
Claims
1. A dynamic resource slicing method for an air-ground integrated vehicle network, which is used in an air-ground integrated vehicle network scenario; the scenario includes multiple base stations, multiple vehicles, and a controller; the base stations are divided into ground base stations and drone base stations; the base stations receive all task requests from vehicles within the coverage area, and are characterized by According to the two types of tasks, delay-sensitive and delay-tolerant, the spectrum resources of each base station are divided into two spectrum resource slices for serving the two types of tasks respectively; the slice time window T is divided into multiple scheduling windows t of equal size; within the slice time window T, the base station allocates spectrum resources according to the slice scheme; Within the scheduling window t, the controller performs workflow scheduling on the received tasks; The controller formulates a spectrum resource segmentation strategy based on the service type and resource requirements of the tasks in the road network, and slices the spectrum resources according to the segmentation strategy; The steps of workflow scheduling are as follows: the vehicle submits a task request to the base station; the base station receives the task and unloads it to the task unloading queue in the controller; The base station processes the task according to the workflow decision of the controller; the base station transmits the processing result of the task back to the vehicle, and the base station uploads the task data to the controller; The spectrum resource segmentation strategy is formulated based on the scheduling data of the historical window, which reduces the calculation pressure of the base station; on the other hand, it is based on the actual situation of the road network to cope with the dynamic changes of traffic; Workflow scheduling needs to comprehensively consider multiple factors, such as task service type, QoS restrictions, the number of allocatable resources in the base station, and queuing information; Dynamically divide spectrum resources of each base station and optimize workflow scheduling to adapt to the differentiation of service types in the network and the dynamic nature of road network traffic; There is task delay in the workflow scheduling process. Task delay is the sum of unloading delay and handover delay. Among them: a. Offloading delay represents the time it takes for task m to be transmitted from vehicle i to base station j. The average time it takes for a task with request type o to be transmitted from a vehicle to a base station is quantized as r i,j,o is the uplink transmission rate of the request type o task generated by vehicle i within the coverage area of base station j, ε i,m,o The set of tasks in the slice window T is expressed as Task m generated by vehicle i contains the delay requirement for task processing completion; The task arrival of a single vehicle is modeled as a Poisson process, and the task arrival received by the base station is also modeled as a Poisson process accordingly; define the binary variable a i,j =1 represents that vehicle i is associated with base station j; the arrival rate of request type o task is expressed as where λ i,o represents the arrival rate of vehicle i generating a request type o task, and the set of ground base stations, drone base stations and vehicles are represented as and The offloading queue only processes one task at a time; the task offloading process is modeled as an M / M / 1 queue model; the entry of the offloading queue is determined by the task arrival rate, and the exit of the offloading queue is determined by the base station transmission; when the queue entry rate is greater than the exit rate, the tasks in the queue will continue to accumulate and cause the queue to overflow; the queue reflects the busyness with service intensity, and the service intensity of the offloading queue with request type o in base station j is defined as Where μ o The average time it takes for a task with request type o to be transmitted from the vehicle to the base station In order to maintain the stability of the unloading queue and prevent queue overflow, formula (6) needs to satisfy The position of the task in the queue is calculated from the head of the queue; the set of the first n tasks in the unloading queue of base station j is represented as Ω j (n); Assuming that task m with request type o generated by vehicle i is unloaded to base station j and its rank in the unloading queue is n, the unloading delay of the task is expressed as b. The handover delay of task m of request type o in base station j back to vehicle i is expressed as The task delay consists of unloading delay and handover delay. The task delay of request type o task m generated by vehicle i is The completion of delay-sensitive tasks requires two conditions to be met: to meet the delay requirement, it must be before the vehicle leaves the coverage area of the base station; then: Definition i,j,m The distance from the edge of base station j when vehicle i generates task m; the maximum travel time within the coverage of base station j when vehicle i generates task m is calculated as Among them, v i represents the driving speed of vehicle i; the constraint on the completion of delay-sensitive tasks is expressed as 2. The dynamic resource slicing method for air-ground integrated vehicle networking according to claim 1 is characterized in that During workflow scheduling: The task offloading queue of the controller includes two queues: delay-sensitive and delay-tolerant. The task processing queue of the base station also has two corresponding queues: delay-sensitive and delay-tolerant. The tasks generated by the vehicle are received by each base station and unloaded to the corresponding task unloading queue in the controller according to the task type; following the principle of first come first served, the controller transfers the task to the base station for processing; after receiving the task, the base station puts the task into the corresponding processing queue and allocates physical resources for processing; The base station hands over the processed results to the vehicle that issued the request.
3. The dynamic resource slicing method for air-ground integrated vehicle networking according to claim 1 is characterized in that The performance optimization within the slice time window T depends on the spectrum resource slice strategy and workflow scheduling strategy. Suppose the spectrum resource slice strategy Workflow scheduling strategy The number of subchannels of base station j in the slice time window T is Define the following binary variables Indicates whether the task processing meets the delay constraint after vehicle i connects to base station j in the tth scheduling window; if the delay requirement can be met, then otherwise When the task is completed under the condition of satisfying the delay constraint, the corresponding benefit is obtained; correspondingly, if the task is not completed, the corresponding loss is generated; Definition 1: In the tth scheduling window, when the task is completed and the delay requirement is met, the total reward U (t) where u j,o ∈(0,1) represents the corresponding profit factor of the task with request type o on base station j; Definition 2: When a task fails to meet the delay requirement in the tth scheduling window, the total loss H is (t) where h j,o ∈(0,1) represents the corresponding loss factor of the task with request type o on base station j; With the goal of maximizing the number of vehicle mission completions, the dynamic RAN slicing problem is modeled as Problem P0 is to coordinate the allocation of spectrum resources of each base station and the workload of the region in an online manner so that the average number of tasks completed by the system within the slicing window T is maximized; constraint a ensures that the number of sub-channels allocated to each base station j is a positive number, constraint b ensures that the number of channels allocated to vehicles by each base station does not exceed the total number of channels it holds, constraint c ensures that each vehicle can only connect to a unique ground base station or drone base station, and constraint d ensures the stability of the queue in the queuing system. At the same time, it also shows that the slicing decision of the spectrum resources and the workflow allocation decision are coupled, that is, the coupling constraint.
4. The dynamic resource slicing method for air-ground integrated vehicle networking according to claim 3 is characterized by: Decompose the problem P0 into spectrum resource slicing sub-problem and workflow scheduling sub-problem; The spectrum resource slicing sub-problem solves the optimal slicing decision by given workflow scheduling decisions; a. Spectrum resource slicing sub-problem at a large time scale The spectrum resource slicing sub-problem is to make scheduling decisions for a given workflow. Under the condition of optimizing resource slicing decision To maximize the number of requests completed by the system, that is The essence of problem P1 is to maximize the number of completed tasks in each slice window; the decision of each slice window is independent and does not affect each other; the traffic flow of adjacent slice windows is similar; then, the controller uses the actual workflow decision of the previous slice window to optimize the resource slice; The controller records the workflow decisions within each slice window and provides a reference for optimizing resource slice decisions for the next slice window to maximize the number of requests completed within each slice window. The above problem P1 corresponds to the following one-shot optimization problem: Problem P2 is solved using the Lagrange multiplier method, and problem P2 is transformed into P3: Given a workflow decision, the controller calculates the specific number of tasks to be processed by each base station. Then, the spectrum resource slicing problem is constructed based on the spectrum resource requirements and delay requirements of the tasks and the spectrum resource holdings of each base station. An optimal spectrum resource slicing solution is obtained by calculating P3. b. Small time scale workflow scheduling sub-problem The workflow scheduling sub-problem is to make decisions on slicing given spectrum resources. Under the condition of To maximize the number of tasks completed by the system; When solving the workflow scheduling subproblem P4, the long-term optimization problem is decomposed into short-term optimization problems within each scheduling window; Workflow scheduling needs to consider the number of tasks waiting to be processed, the spectrum resource requirements of each task, the waiting time and delay requirements, and the queue processing information of each base station; the controller needs the information of each task in the unloading queue, the spectrum resource scheduling of each base station, and the task information in the internal processing queue; the controller allocates the tasks in the unloading queue according to the first-come-first-served strategy, and the specific workflow scheduling algorithm processes the first n task set Ω(n) in the unloading queue in the controller to obtain the optimal allocation decision for each task; Among all base stations, if for a base station: Determine whether the processing time of the task in each processing queue meets the delay requirement; if so, remove the base station from the calculation process; Determine whether the remaining spectrum resources of each base station can meet the current task requirements; if so, remove the base station from the calculation process; Determine whether the maximum travel time of the current task in each base station meets the delay requirement; if so, remove the base station from the calculation process; c. Integrate the spectrum resource slicing sub-problem with the workflow scheduling sub-problem to solve: First, the system divides the length of the slice window according to the task information in the historical data; After the slice window is determined, the workflow scheduling decision within the T-1th slice window is made As a known condition for solving problem P3, and solving the resource slicing decision Workflow scheduling decision for the first slice window Given by historical data; Divide the slice window T into multiple scheduling windows of equal size T = {1,...,t,...}. In each scheduling window, make resource slice decisions As a known condition for solving problem P4, the workflow scheduling decision within each scheduling window is obtained; Each base station processes tasks according to workflow scheduling decisions; At the end of the last scheduling window, the workflow scheduling decisions within each scheduling window are integrated into the workflow scheduling decision of slice window T And recorded as historical data for use in the T+1th slice window.
5. The dynamic resource slicing method for air-ground integrated vehicle networking according to claim 1 is characterized in that The communication method between the base station and the vehicle is as follows: the base station allocates spectrum resources in the same resource slice to the associated vehicles in an orthogonal manner according to the task request type.
