A method for collaborative unloading between air and ground

Abstract

This invention discloses a collaborative air-ground offloading method that leverages the directed acyclic graph of applications to perform refined offloading, minimizing system task processing latency. It proposes an optimal power allocation algorithm and a heuristic task offloading algorithm, which iterate over each other to achieve an optimal task offloading and collaborative communication solution. Compared with existing mechanisms, this method can effectively reduce the average latency of in-vehicle tasks and improve the quality of service for vehicle users.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle networking, and particularly relates to an air-ground collaborative unloading method. Background Art

[0002] As one of the representative scenarios of 5G, VANETs (Vehicular Ad Hoc Networks) can provide users with a safer and more enjoyable driving experience by acquiring, managing, and calculating large-scale dynamic data information from people, vehicles, objects, and the environment through network connectivity and information exchange between V2X (X can be people, vehicles, networks, infrastructure, etc.).

[0003] With the development of the Internet of Vehicles (IoV), the application of new in-vehicle terminals is placing increasingly stringent demands on communication quality and latency. These applications often require powerful computing power to process complex data and ensure low latency. However, resource-limited on-board units (IoVs) cannot simultaneously meet the demands of high computational load and low latency, making task offloading essential. In recent years, cloud computing has been introduced into IoV systems. When a user selects an application service, the on-board device transmits the relevant task data to the cloud server, which processes the task based on its powerful computing resources, thereby eliminating the hardware limitations of the mobile terminal. However, due to the limited communication transmission capacity of the 4G network and the long transmission distance to the cloud, the method of aggregating task data to the cloud for processing not only wastes network bandwidth and has a large latency, but also causes service unavailability in the event of network jitter or transmission interruption, reducing user service quality.

[0004] In response to the above problems, Mobile Edge Computing (MEC) came into being. It sinks computing, storage, processing and other functions from the centralized cloud platform to the edge side of the wireless network, which can reduce communication delays considerably. Task offloading in mobile edge computing is a key issue that needs to be discussed in this invention. In the drone-assisted Internet of Vehicles (referred to as the air-ground collaborative Internet of Vehicles in this invention), some existing studies adopt a binary offloading strategy, treating the task as a whole, and executing it all locally or migrating it all to the edge server for execution. Although this method is simple and easy to apply, it does not take into account the parallelism between subtasks for refined offloading and does not fully utilize the computing resources of the system. There are also many existing studies that do not consider the impact of communication spectrum and power allocation, and under different conditions such as channel state information feedback cycle and vehicle driving speed, the channel transmission capacity varies greatly, which affects the offloaded data transmission delay. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides an air-ground collaborative offloading method. This method leverages the directed acyclic graph of applications to perform refined offloading, minimizing system task processing latency. It proposes an optimal power allocation algorithm and a heuristic task offloading algorithm, which iterate over each other to achieve an optimal task offloading and collaborative communication solution. Compared with existing mechanisms, this invention can effectively reduce the average latency of in-vehicle tasks and improve the quality of service for vehicle users.

[0006] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0007] Step 1: Build an air-ground collaborative vehicle network model consisting of N mobile vehicles and one drone base station;

[0008] Step 2: Based on the air-ground collaborative vehicle network model, build the system communication model, energy consumption model, and delay model;

[0009] Step 3: Based on the air-ground collaborative vehicle network model, a fixed task offloading strategy is adopted to propose an optimal resource allocation algorithm;

[0010] Step 4: Based on the air-ground collaborative vehicle network model, a fixed resource allocation scheme is proposed to propose a task offloading algorithm;

[0011] Step 5: Based on the proposed optimal resource allocation algorithm and task offloading algorithm, an iterative framework is constructed to obtain the optimal task offloading and collaborative communication solution.

[0012] Furthermore, the step 1 specifically includes:

[0013] Step 1-1: Build an air-ground cooperative vehicle network model consisting of N mobile vehicles and one drone base station. The N vehicles require V2I communication, denoted as CUE; and M pairs of vehicles share V2V information through end-to-end communication, denoted as DUE.

[0014] Step 1-2: Each vehicle has an application to complete, which is divided into multiple subtasks, represented by a directed acyclic graph. Each subtask can be executed locally or offloaded to the edge server for execution; V i ={1,2,...,j,...,V i} represents a set of subtasks of the application to be completed on vehicle i;

[0015] Furthermore, the step 2 specifically includes:

[0016] Step 2-1: The channel gain between the nth CUE and the base station is expressed as g n,B :

[0017] g n,B =|h n,B |2 α n,B , (1)

[0018] Among them, h n,B is the small-scale fast fading component, assuming that Independent and identically distributed; α n,B Capture all large-scale fading effects;

[0019] Similarly, the channel gain between the mth V2V pair is defined as g m , the interference channel gain from the mth DUE to the base station is And the interference channel from the nth CUE to the mth DUE is g n,m ;

[0020] For links with inaccurate channel state information, i.e., g n,B and A first-order Gauss-Maldives process is used to simulate the channel variation over a period T:

[0021]

[0022] Among them, h and represent the current and previous channel states respectively; e is the channel difference term; ε=J0(2πf d T), where J0(.) is the zero-order Bessel function of the first kind, f d =vf c / c is the maximum Doppler frequency, c = 3 × 10 8 m / s, v is the vehicle speed, f c is the carrier frequency;

[0023] Step 2-2: The signal-to-noise ratio of the nth CUE and the mth DUE is expressed as:

[0024]

[0025]

[0026] in, denote the transmission power of the nth CUE and the mth DUE, σ 2 is the noise power, ρ n,m =1 means that the mth DUE chooses to reuse the spectrum of the nth CUE, otherwise ρ n,m =0;

[0027] The transmission rate of the nth CUE channel corresponding to vehicle i is expressed as:

[0028]

[0029] Among them, B irepresents the bandwidth between vehicle i and the edge server;

[0030] Step 2-3: Define an uninstall decision set Indicates that the jth subtask on vehicle i is executed locally, It means that it is offloaded to the edge server for execution; the energy consumption of vehicle i is expressed as:

[0031] Cost i =Cost c +Cost t , (6)

[0032] Cost c is the local computing energy consumption of the vehicle, which is calculated as follows:

[0033]

[0034] Among them, I X Is an indicator function. When condition X is met, I X is equal to 1, otherwise equal to 0; i is the computational consumption per unit time of vehicle i;

[0035] Cost t It is the energy consumption of vehicle data offloading transmission, which mainly includes two parts: uploading task consumption and data conversion consumption between subtasks. Its calculation formula is:

[0036]

[0037] in, is the transmission power of the nth CUE corresponding to vehicle i, Size i,j is the data size of the corresponding subtask, data j'j represents the amount of data exchange between subtasks j and j;

[0038] Step 2-4: As each subtask is executed either locally or on an edge server, the application G on vehicle i i The actual earliest completion time of subtask j is expressed as:

[0039]

[0040] in, is the execution time of subtask j, where k = 0 represents local execution and k = 1 represents offloading to the edge server for execution;

[0041] The actual execution start time EST of subtask j on k A (i,j,k) is expressed as:

[0042] EST A (i,j,k)=max{avail{0∪[k]},EST T (i,j,k)}, (10)

[0043]

[0044] Among them, EST T (i, j, k) is the ideal earliest start time for executing subtask j on k; avail{0∪[k]} represents the earliest time when server k is ready to start executing subtask j, pred(j) is the set of direct predecessors of subtask j, and C jj' is the data conversion time between subtasks, which is defined as:

[0045]

[0046] Through continuous iteration, after all subtasks are scheduled, the minimum execution delay of the vehicle i application is obtained as:

[0047]

[0048] in, represents the last subtask of vehicle i;

[0049] Step 2-5: According to formula (13), a joint optimization problem based on task offloading decision, spectrum and power allocation is proposed. The specific expression is:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in is the minimum throughput indicator of CUE, is the minimum SINR required by DUE to ensure link reliability, Pr{} evaluates the probability of input, p0 is the tolerable outage probability, and are the maximum transmission powers of CUE and DUE respectively.

[0061] Furthermore, the step 3 specifically includes:

[0062] Step 3-1: Decouple the joint optimization problem into two sub-problems: resource allocation problem and offloading decision problem;

[0063] Step 3-2: Optimize resource allocation with a fixed task offloading strategy. The optimization problem is formulated as:

[0064]

[0065] st formula 14 (a)-14 (f)

[0066] For each CUE and DUE pair, the maximum capacity of CUE is:

[0067]

[0068] At the same time, all constraints are met, and the formula is expressed as:

[0069]

[0070] st formula 14(b)-14(d)

[0071] The SINR received at the mth DUE is simplified to in and X and Y are two independent exponential random variables with unit mean;

[0072] Step 3-3: Evaluate the reliability of the V2V link from two situations:

[0073] Case 1: When hour:

[0074]

[0075] Case 2: When hour:

[0076]

[0077] For the two cases, hour:

[0078]

[0079] when hour:

[0080]

[0081] in, and The transmission power of DUE and CUE at the critical point of the two cases are respectively, according to The relationship between the size of A and the feasible region of formula (17) is divided into two parts, and the upper boundaries of these two regions intersect at Located on the dividing line Above; combined we can get:

[0082]

[0083]

[0084] function and respectively and Keep within the value range and The monotonically increasing relationship between them; at the same time, the capacity of CUE is equivalent to along with increases with the increase of Therefore, the optimal solution is located at the upper boundary of the feasible region, and the optimal power allocation solution is given by and The relative size of and their intersection with the boundary line determine the optimal power allocation solution of equation (17):

[0085]

[0086]

[0087] and By implicit function and Calculated;

[0088] and By implicit function and Calculated;

[0089] Step 3-4: Calculate the optimal power allocation results for all single CUE and DUE pairs according to the above method. Substituting into formula (16) we get exclude For the combination pairs that cannot meet the minimum capacity requirement of CUE, all possible combinations of multiplexing pairs are evaluated, and then the Hungarian algorithm is used to find the optimal resource allocation method.

[0090] Furthermore, the step 4 specifically includes:

[0091] Step 4-1: Calculate the task offloading strategy under fixed spectrum and power allocation; sort the subtasks based on the given directed acyclic graph. The ranking of all subtasks is expressed as:

[0092]

[0093] in, is the average execution time of subtask j, is the average data transmission time between subtasks; Succ(j) is the set of subtasks directly inherited by subtask j, that is, the set of subtasks that need to be executed after this subtask;

[0094] The priority of a subtask is defined as:

[0095]

[0096] Among them, α i , β i are the weight coefficients of vehicle i regarding computational overhead and energy consumption, respectively;

[0097] Calculate the rank(i,j) of all subtasks of all applications;

[0098] Step 4-2: Find the highest-priority unscheduled subtasks in all applications to form a candidate scheduling set M, calculate the Prio(i, j) of all subtasks in set M, and find the subtask j with the highest priority. one By comparing EFT(i,j one ) decides whether to assign subtasks to edge servers or local servers;

[0099] Step 4-3: Repeat step 4-2 until all subtasks are scheduled and output the overall offloading strategy.

[0100] Furthermore, the step 5 specifically includes:

[0101] Step 5-1: Initialize the spectrum, power allocation, and task offloading strategies, decompose the joint optimization problem into a resource allocation subproblem and a computation offloading subproblem, and optimize each subproblem using the corresponding algorithm while keeping the variables of the other subproblem fixed.

[0102] Step 5-2: Based on the initial value, calculate the result of the first iteration and use it as the starting value for the next iteration. Repeat the iteration until the algorithm converges and the optimal task offloading and cooperative communication solution is obtained.

[0103] The beneficial effects of the present invention are as follows:

[0104] This invention addresses the problems of existing technologies and, in light of the current situation of limited local vehicle computing power and spectrum resource shortages, proposes an air-ground collaborative offloading method. This method fully utilizes the directed acyclic graph of applications for refined offloading, aiming to minimize system task processing latency. It also proposes an optimal power allocation algorithm and a heuristic task offloading algorithm, which iterate over each other to achieve an optimal task offloading and collaborative communication solution. Compared with existing mechanisms, this invention can effectively reduce vehicle-based task processing latency and improve the quality of service for vehicle users. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] Figure 1 It is a flow chart of the method of the present invention.

[0106] Figure 2 This is a performance comparison chart comparing the average delay of the potential game-based offloading algorithm (PGOA) provided by an embodiment of the present invention.

[0107] Figure 3 This is a performance comparison chart comparing energy consumption with local computing provided by an embodiment of the present invention.

[0108] Figure 4 This is a comparison diagram of the relationship between the total V2I link capacity, the CSI feedback period, and the vehicle speed provided by an embodiment of the present invention.

[0109] Figure 5 This is a comparison chart of the relationship between the average delay of a vehicle application and the number of split subtasks provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0110] The present invention will be further described below with reference to the accompanying drawings and examples.

[0111] This paper addresses the challenges of existing technologies and addresses the current situation of limited local vehicle computing power and spectrum resource shortages. It proposes a collaborative air-ground offloading method. This method leverages the directed acyclic graph of applications to perform refined offloading, minimizing system task processing latency. It also proposes an optimal power allocation algorithm and a heuristic task offloading algorithm, which iterate over each other to achieve an optimal task offloading and collaborative communication solution.

[0112] like Figure 1 As shown, the air-ground collaborative unloading method provided by the embodiment of the present invention includes the following steps:

[0113] S101: Build an air-ground collaborative vehicle network model consisting of N mobile vehicles and one drone base station;

[0114] S102: Based on the air-ground collaborative vehicle network model, build the system communication model, energy consumption model, and delay model;

[0115] S103: Based on the air-ground collaborative vehicle network model and fixed task offloading strategy, an optimal resource allocation algorithm is proposed;

[0116] S104: Based on the air-ground collaborative vehicle network model and fixed resource allocation scheme, a task offloading algorithm is proposed;

[0117] S105: Based on the proposed optimal resource allocation algorithm and task offloading algorithm, an iterative framework is constructed to obtain an optimal task offloading and collaborative communication solution.

[0118] The air-ground coordinated unloading method provided by the embodiment of the present invention includes the following steps:

[0119] The first step is to build an air-ground collaborative vehicle network model consisting of N mobile vehicles and one drone base station. The specific steps are as follows:

[0120] In step (1.1), a collaborative air-ground vehicle network model consisting of N mobile vehicles and one UAV base station is constructed, where the N vehicles require high-capacity V2I communication, denoted as CUE; and M pairs of vehicles share V2V information through end-to-end communication, denoted as DUE.

[0121] In step (1.2), each vehicle has an application to complete, which can be divided into multiple subtasks, represented by a directed acyclic graph. Each subtask can be executed locally or offloaded to the edge server. i ={1,2,...,j,...,V i} represents a set of subtasks of the application to be completed on vehicle i;

[0122] The second step is to build a system communication model, energy consumption model, and delay model based on the air-ground collaborative vehicle network model. The specific steps are as follows:

[0123] In step (2.1), the channel gain between the nth CUE and the base station is expressed as g n,B .

[0124] g n,B =|h n,B | 2 α n,B ,(S.1)

[0125] Among them, h n,B is the small-scale fast fading component, assuming that Independent and identically distributed; α n,B Captures all large-scale fading effects, including path loss and shadowing.

[0126] Similarly, the channel gain between the mth V2V pair is defined as g m , the interference channel gain from the mth DUE to the base station is And the interference channel from the nth CUE to the mth DUE is g n,m .

[0127] For links with inaccurate channel state information, i.e., g n,B and A first-order Gaussian-Maldives process is used to simulate the channel variation over a period T.

[0128]

[0129] Among them, h and Represent the current and previous channel states respectively; e is the channel difference term, independent of ε=J0(2πf d T), where J0(.) is the zero-order Bessel function of the first kind, f d =vf c / c is the maximum Doppler frequency, c = 3 × 10 8 m / s, v is the vehicle speed, f c is the carrier frequency;

[0130] In step (2.2), the signal-to-noise ratio of the nth CUE and the mth DUE is expressed as:

[0131]

[0132]

[0133] in, Represent the transmission power of the nth CUE and the mth DUE respectively. 2 is the noise power, ρ n,m =1 means that the mth DUE chooses to reuse the spectrum of the nth CUE, otherwise ρ n,m =0.

[0134] The transmission rate of the nth CUE channel corresponding to vehicle i is expressed as:

[0135]

[0136] Among them, B i represents the bandwidth between vehicle i and the edge server;

[0137] Step (2.3), define an uninstall decision set Indicates that the jth subtask on vehicle i is executed locally, It means that the execution is offloaded to the edge server. The energy consumption of vehicle i can be expressed as:

[0138] Cost i =Cost c +Cost t ,(S.6)

[0139] Cost c is the local computing energy consumption of the vehicle, which is calculated as follows:

[0140]

[0141] I X Is an indicator function. When condition X is met, I X is equal to 1, otherwise equal to 0. i is the computational cost per unit time of vehicle i.

[0142] Cost t It is the energy consumption of vehicle data offloading transmission, which mainly includes two parts: uploading task consumption and data conversion consumption between subtasks. Its calculation formula is:

[0143]

[0144] in, is the transmission power of the nth CUE corresponding to vehicle i, Size i,j is the data size of the corresponding subtask, data j'j represents the amount of data exchange between subtasks j and j;

[0145] In step (2.4), since each subtask is executed either locally or on the edge server, the application G on vehicle i i The actual earliest completion time of subtask j can be expressed as:

[0146]

[0147] in, is the execution time of subtask j, where k = 0 represents local execution and k = 1 represents offloading to the edge server for execution.

[0148] The actual execution start time EST of subtask j on k A (i,j,k) is expressed as:

[0149] EST A (i,j,k)=max{avail{0∪[k]},EST T(i,j,k)},(S.10)

[0150]

[0151] Among them, EST T (i, j, k) is the ideal earliest start time for executing subtask j on k. avail{0∪[k]} represents the earliest time when server k is ready to start executing subtask j, pred(j) is the set of direct predecessor subtasks of subtask j, and C jj' is the data conversion time between subtasks, which is defined as:

[0152]

[0153] Through continuous iteration, after all subtasks are scheduled, the minimum execution delay of the vehicle i application can be obtained as:

[0154]

[0155] in, represents the last subtask of vehicle i;

[0156] In step (2.5), according to formula (S.13), a joint optimization problem based on task offloading decision, spectrum and power allocation is proposed. The specific expression is:

[0157]

[0158]

[0159]

[0160]

[0161]

[0162]

[0163]

[0164]

[0165]

[0166]

[0167] in is the minimum throughput indicator of CUE, is the minimum SINR required by DUE to ensure link reliability. Pr{} evaluates the probability of input and p0 is the tolerable outage probability. and are the maximum transmission powers of CUE and DUE respectively;

[0168] The third step is to propose an optimal resource allocation algorithm based on the air-ground collaborative vehicle network model and fixed task offloading strategy. The specific steps are as follows:

[0169] In step (3.1), the joint optimization problem proposed by this method is decoupled into two sub-problems, namely the resource allocation problem and the offloading decision problem. Accordingly, an optimal resource allocation algorithm and a heuristic task offloading algorithm are proposed.

[0170] In step (3.2), the fixed task offloading strategy optimizes resource allocation to meet the different requirements of different vehicle links, namely, large capacity of V2I connection and high reliability of V2V connection, while limiting the minimum capacity of each CUE to ensure that the minimum user service quality can be provided. The optimization problem is formulated as:

[0171]

[0172] st formula (S.14a)-(S.14f)

[0173] Now we decouple the problem and focus on each CUE and DUE pair to maximize the capacity of CUE:

[0174]

[0175] At the same time, all constraints are met, and the formula is expressed as:

[0176]

[0177] st formula (S.14b)-(S.14d)

[0178] The SINR received at the mth DUE can be simplified to in and X and Y are two independent exponential random variables with unit mean;

[0179] Step (3.3) evaluates the reliability of the V2V link from two situations:

[0180] Case 1: When hour

[0181]

[0182] Case 2: When hour

[0183]

[0184] For the two cases, hour,

[0185]

[0186] when hour,

[0187]

[0188] in, and The transmission power of DUE and CUE at the critical point of the two cases are respectively, according to The relationship between the size of and A divides the feasible region of (S.17) into two parts, and the upper boundaries of these two regions intersect at It is located on the dividing line Above, we can get:

[0189]

[0190]

[0191] function and respectively and Keep within the value range and The monotonically increasing relationship between them. At the same time, the capacity of CUE is equivalent to along with increases with the increase of Therefore, the optimal solution must be located at the upper boundary of the feasible region, and the optimal power allocation solution is given by and The relative sizes of and their intersections with the boundary lines determine the optimal power allocation solution of (S.17):

[0192]

[0193]

[0194] and By implicit function and Calculated;

[0195] and By implicit function and Calculated;

[0196] Step (3.4): Calculate the optimal power allocation results for all single CUE and DUE pairs according to the above method. Substituting into (S.16) we get exclude For the combination pairs that cannot meet the minimum capacity requirement of CUE, all possible combinations of multiplexing pairs are evaluated, and then the Hungarian algorithm is used to find the optimal resource allocation method;

[0197] The fourth step is to propose a task offloading algorithm based on the air-ground collaborative vehicle network model and a fixed resource allocation scheme. The specific steps are as follows:

[0198] In step (4.1), the task offloading strategy is calculated under fixed spectrum and power allocation. Based on the given directed acyclic graph, the subtasks are sorted and the levels of all subtasks can be expressed as:

[0199]

[0200] in, is the average execution time of subtask j, is the average data transmission time between subtasks. Succ(j) is the set of subtasks directly inherited by subtask j, that is, the set of subtasks that need to be executed after this subtask. Further considering energy consumption, subtasks with higher computational overhead and higher local energy consumption are prioritized for offloading. The priority of subtasks is defined as:

[0201]

[0202] Among them, α i , β i They are the weight coefficients of vehicle i regarding computational overhead and energy consumption, as long as α is guaranteed i +β i =1, different coefficient proportions can be set according to different scenario requirements.

[0203] First, calculate the rank (i, j) of all subtasks of all applications;

[0204] Step (4.2): Find the highest-priority unscheduled subtasks in all applications to form a candidate scheduling set M, calculate the Prio(i, j) of all subtasks in set M, and find the subtask j with the highest priority. one By comparing EFT(i,j one ) decides whether to assign subtasks to edge servers or local servers;

[0205] Step (4.3), repeat step (4.2) until all subtasks are scheduled and the overall offloading strategy is output;

[0206] The fifth step is to build an iterative framework based on the proposed optimal resource allocation algorithm and task offloading algorithm to obtain the optimal task offloading and collaborative communication solution. The specific steps are as follows:

[0207] In step (5.1), the spectrum, power allocation, and task offloading strategies are initialized, and the joint optimization problem is decomposed into a resource allocation subproblem and a computation offloading subproblem. Each subproblem is optimized using the corresponding algorithm while the variables of the other subproblem are fixed.

[0208] In step (5.2), the result of the first iteration is calculated based on the initial value, and it is used as the starting value of the next iteration. The iteration is repeated until the algorithm converges and the optimal task offloading and cooperative communication scheme is obtained.

[0209] The technical effects of the present invention are described in detail below with reference to simulations.

[0210] This experiment simulates an air-ground collaborative offloading method to verify the superiority of the proposed method. The specific steps are as follows: Following the highway scenario in 3GPP TR 36.885, a multi-lane highway with a single cell radius of 500m is modeled. Vehicle positions are distributed according to a spatial Poisson process, with density adjusted based on vehicle speed. The maximum transmission power is 23dBm. CUE minimum capacity 0.5bps / Hz, DUE minimum signal-to-noise ratio is 5dB, and the DUE reliability probability p0 is 10 -3 .

[0211] The average delay, energy consumption and total capacity of the V2I link under different conditions are statistically analyzed to obtain the simulation results.

[0212] The present invention is compared with the potential game-based offloading algorithm (PGOA) and local computing, such as Figure 2 、 Figure 3 The relationship between the total V2I link capacity, CSI feedback cycle and vehicle speed is shown in Figure 4 As shown in the figure, the relationship between the average delay and the number of subtasks is as follows Figure 5 shown.

[0213] In summary, the embodiment of the present invention provides an air-ground collaborative unloading method. In response to the problems existing in the existing technology, combined with the current situation of limited local computing power and shortage of spectrum resources in vehicles, an air-ground collaborative unloading method is proposed. This method makes full use of the directed acyclic graph of the application to perform refined unloading to achieve the goal of minimizing the system task processing delay. An optimal power allocation algorithm and a heuristic task unloading algorithm are proposed. The two algorithms iterate each other to obtain the optimal task unloading and collaborative communication scheme. Compared with existing mechanisms, the present invention can effectively reduce the average delay of vehicle-mounted tasks and improve the quality of service for vehicle users.

Claims

1. A method for air-ground collaborative unloading, characterized in that: The steps include: Step 1: Build an air-ground collaborative vehicle network model consisting of N mobile vehicles and one drone base station; Step 1-1: Build an air-ground cooperative vehicle network model consisting of N mobile vehicles and one drone base station. The N vehicles require V2I communication, denoted as CUE; and M pairs of vehicles share V2V information through end-to-end communication, denoted as DUE. Step 1-2: Each vehicle has an application to complete, which is divided into multiple subtasks, represented by a directed acyclic graph. Each subtask can be executed locally or offloaded to the edge server for execution; V i ={1,2,...,j,...,V i } represents a set of subtasks of the application to be completed on vehicle i; Step 2: Based on the air-ground collaborative vehicle network model, build the system communication model, energy consumption model, and delay model; Step 2-1: The channel gain between the nth CUE and the base station is expressed as g n,B : g n,B =|h n,B | 2 a n,B , (1) Among them, h n,B is the small-scale fast fading component, assuming that Independent and identically distributed; α n,B Capture all large-scale fading effects; Similarly, the channel gain between the mth V2V pair is defined as g m , the interference channel gain from the mth DUE to the base station is And the interference channel from the nth CUE to the mth DUE is g n,m ; For links with inaccurate channel state information, i.e., g n,B and A first-order Gauss-Maldives process is used to simulate the channel variation over a period T: Among them, h and represent the current and previous channel states respectively; e is the channel difference term; ε=J0(2πf d T), where J0(.) is the zero-order Bessel function of the first kind, f d =vf c / c is the maximum Doppler frequency, c = 3 × 10 8 m / s, v is the vehicle speed, f c is the carrier frequency; Step 2-2: The signal-to-noise ratio of the nth CUE and the mth DUE is expressed as: in, denote the transmission power of the nth CUE and the mth DUE, σ 2 is the noise power, ρ n,m =1 means that the mth DUE chooses to reuse the spectrum of the nth CUE, otherwise ρ n,m =0; The transmission rate of the nth CUE channel corresponding to vehicle i is expressed as: Among them, B i represents the bandwidth between vehicle i and the edge server; Step 2-3: Define an uninstall decision set Indicates that the jth subtask on vehicle i is executed locally, It means that it is offloaded to the edge server for execution; the energy consumption of vehicle i is expressed as: Cost i =Cost c +Cost t , (6) Cost c is the local computing energy consumption of the vehicle, which is calculated as follows: Among them, I X Is an indicator function. When condition X is met, I X is equal to 1, otherwise equal to 0; i is the computational consumption per unit time of vehicle i; Cost t It is the energy consumption of vehicle data offloading transmission, which mainly includes two parts: uploading task consumption and data conversion consumption between subtasks. Its calculation formula is: in, is the transmission power of the nth CUE corresponding to vehicle i, Size i,j is the data size of the corresponding subtask, data j'j represents the amount of data exchange between subtasks j and j; Step 2-4: As each subtask is executed either locally or on an edge server, the application G on vehicle i i The actual earliest completion time of subtask j is expressed as: in, is the execution time of subtask j, where k = 0 represents local execution and k = 1 represents offloading to the edge server for execution; The actual execution start time EST of subtask j on k A (i,j,k) is expressed as: EAST A (i,j,k)=max{avail{0∪[k]},EST T (i,j,k)}, (10) Among them, EST T (i, j, k) is the ideal earliest start time for executing subtask j on k; avail{0∪[k]} represents the earliest time when server k is ready to start executing subtask j, pred(j) is the set of direct predecessors of subtask j, and C jj' is the data conversion time between subtasks, which is defined as: Through continuous iteration, after all subtasks are scheduled, the minimum execution delay of the vehicle i application is obtained as: in, represents the last subtask of vehicle i; Step 2-5: According to formula (13), a joint optimization problem based on task offloading decision, spectrum and power allocation is proposed. The specific expression is: in is the minimum throughput indicator of CUE, is the minimum SINR required by DUE to ensure link reliability, Pr{} evaluates the probability of input, p0 is the tolerable outage probability, and are the maximum transmission powers of CUE and DUE respectively; Step 3: Based on the air-ground collaborative vehicle network model, a fixed task offloading strategy is adopted to propose an optimal resource allocation algorithm; Step 3-1: Decouple the joint optimization problem into two sub-problems: resource allocation problem and offloading decision problem; Step 3-2: Optimize resource allocation with a fixed task offloading strategy. The optimization problem is formulated as: st formula 14 (a)-14 (f) For each CUE and DUE pair, the maximum capacity of CUE is: At the same time, all constraints are met, and the formula is expressed as: st formula 14(b)-14(d) The SINR received at the mth DUE is simplified to in and X and Y are two independent exponential random variables with unit mean; Step 3-3: Evaluate the reliability of the V2V link from two situations: Case 1: When hour: Case 2: When hour: For the two cases, hour: when hour: in, and The transmission power of DUE and CUE at the critical point of the two cases are respectively, according to The relationship between the size of A and the feasible region of formula (17) is divided into two parts, and the upper boundaries of these two regions intersect at Located on the dividing line Above; combined we can get: function and respectively and Keep within the value range and The monotonically increasing relationship between them; at the same time, the capacity of CUE is equivalent to along with increases with the increase of Therefore, the optimal solution is located at the upper boundary of the feasible region, and the optimal power allocation solution is given by and The relative size of and their intersection with the boundary line determine the optimal power allocation solution of equation (17): and By implicit function and Calculated; and By implicit function and Calculated; Step 3-4: Calculate the optimal power allocation results for all single CUE and DUE pairs according to the above method. Substituting into formula (16) we get exclude For the combination pairs that cannot meet the minimum capacity requirement of CUE, all possible combinations of multiplexing pairs are evaluated, and then the Hungarian algorithm is used to find the optimal resource allocation method; Step 4: Based on the air-ground collaborative vehicle network model, a fixed resource allocation scheme is proposed to propose a task offloading algorithm; Step 5: Based on the proposed optimal resource allocation algorithm and task offloading algorithm, an iterative framework is constructed to obtain the optimal task offloading and collaborative communication solution.

2. The air-ground coordinated unloading method according to claim 1, characterized in that: The step 4 specifically includes: Step 4-1: Calculate the task offloading strategy under fixed spectrum and power allocation; sort the subtasks based on the given directed acyclic graph. The ranking of all subtasks is expressed as: in, is the average execution time of subtask j, is the average data transmission time between subtasks; Succ(j) is the set of subtasks directly inherited by subtask j, that is, the set of subtasks that need to be executed after this subtask; The priority of a subtask is defined as: Among them, α i , β i are the weight coefficients of vehicle i regarding computational overhead and energy consumption, respectively; Calculate the rank(i,j) of all subtasks of all applications; Step 4-2: Find the highest-priority unscheduled subtasks in all applications to form a candidate scheduling set M, calculate the Prio(i, j) of all subtasks in set M, and find the subtask j with the highest priority. one By comparing EFT(i,j one ) decides whether to assign subtasks to edge servers or local servers; Step 4-3: Repeat step 4-2 until all subtasks are scheduled and output the overall offloading strategy.

3. The air-ground coordinated unloading method according to claim 2, characterized in that: The step 5 specifically includes: Step 5-1: Initialize the spectrum, power allocation, and task offloading strategies, decompose the joint optimization problem into a resource allocation subproblem and a computation offloading subproblem, and optimize each subproblem using the corresponding algorithm while keeping the variables of the other subproblem fixed. Step 5-2: Based on the initial value, calculate the result of the first iteration and use it as the starting value for the next iteration. Repeat the iteration until the algorithm converges and the optimal task offloading and cooperative communication solution is obtained.

Images