Multi-vehicle priority task unloading resource allocation method and system based on IMOEA / D

Abstract

The invention relates to the technical field of vehicle task unloading, in particular to a multi-vehicle priority task unloading resource allocation method and system based on IMOEA / D. Tasks of all task vehicles are divided into four priorities, corresponding communication resources are allocated to the tasks of different priorities of the vehicles, and computing resources are allocated to computing nodes; determining a vehicle task unloading mode according to the task information, and establishing a time delay model and an energy consumption model of tasks with different priorities in task transmission and task execution stages; and designing a decomposition-based multi-objective optimization algorithm to solve an optimal task unloading decision. According to the method, the parallel computing relationship among the computing nodes is considered, and a multi-vehicle multi-task unloading strategy can be generated.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle task unloading, and in particular to a multi-vehicle priority task unloading resource allocation method and system based on IMOEA / D. Background Art

[0002] With the development of artificial intelligence and 5G technology, in-vehicle applications are becoming increasingly abundant and have high requirements for latency, such as autonomous driving and virtual reality. Resource-constrained vehicle computing units face huge challenges. Cloud servers can provide vehicles with a large amount of computing resources, but due to the high transmission latency of communication between vehicles and cloud servers, it is difficult to meet the low latency requirements of tasks in the Internet of Vehicles environment by relying solely on cloud servers. As a supplement to cloud computing, fog computing can provide users with low-latency distributed services at a location closer to the vehicle. Introducing fog computing into the Internet of Vehicles environment can effectively alleviate the problem that cloud servers cannot meet the low latency requirements of tasks. However, in the face of dense traffic, the limited number of fog nodes cannot guarantee the service quality of multiple vehicles. Vehicle fog computing (VFC) is a solution that can solve the problem of insufficient computing resources in the Internet of Vehicles. This solution directly connects the task vehicle with the idle vehicles in the environment through V2V (vehicle to vehicle) and uses the computing resources of the idle vehicles to perform tasks. Therefore, the cloud fog Internet of Vehicles system composed of cloud computing, fog computing and vehicle fog computing can provide high-quality services for vehicles.

[0003] Researchers have proposed a variety of solutions to the problem of computing resource allocation and task offloading in the cloud-fog vehicle network environment. Some studies use game theory to obtain the task offloading strategy of the vehicle. Most existing studies assume that the vehicle is stationary or the transmission link state is unchanged, which is inconsistent with the actual situation. Therefore, it is necessary to design a more realistic model to describe the environmental state and transmission link state of the vehicle. In addition, although some existing studies have jointly optimized the task offloading and resource allocation problems, the high complexity of the proposed models and algorithms cannot guarantee the timeliness of solving the task offloading strategy and resource allocation plan. Most of the current studies only consider providing offloading services for the tasks of a single vehicle and ignore the parallel computing relationship of each computing node. If the offloading strategy is still generated for each vehicle in a scene with large traffic volume and dense streets, it will consume a lot of communication resources and computing resources. Summary of the invention

[0004] The present invention aims to solve the problem that the current unloading scheme usually only considers providing unloading services for the tasks of a single vehicle and ignores the parallel computing relationship of each computing node. A multi-vehicle priority task unloading resource allocation method and system based on IMOEA / D are proposed, which considers the parallel computing relationship between each computing node and can generate a multi-vehicle multi-task unloading strategy.

[0005] To achieve the above purpose, the technical solution adopted is:

[0006] A multi-vehicle priority task unloading resource allocation method based on IMOEA / D, comprising:

[0007] Divide the tasks of all mission vehicles into four priorities, and allocate corresponding communication resources to the tasks of different priorities of the vehicles and computing resources to each computing node;

[0008] The vehicle task unloading mode is determined according to the task information, and the delay model and energy consumption model of tasks with different priorities in the task transmission and task execution stages are established;

[0009] A decomposition-based multi-objective optimization algorithm is designed to solve the optimal task offloading decision.

[0010] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, the priority value of the task is determined by three indicators: the deadline of the task, the data volume of the task and the calculation amount of the task. The smaller the priority value, the higher the priority level of the task; after obtaining the priority values ​​of all tasks, all tasks are arranged in ascending order according to the priority values, and the tasks are divided into four priorities.

[0011] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, allocating corresponding communication resources for tasks of different priorities of vehicles includes:

[0012] First, the bandwidth allocated when the task vehicle communicates with the fog node is expressed as:

[0013]

[0014] In the formula, Br j v2f is the bandwidth ratio of the mission vehicle, N represents the number of mission vehicles, B v2f is the communication bandwidth between vehicles and fog nodes;

[0015] Next, the bandwidth allocated to the task vehicle is reallocated; the task with priority I on the task vehicle occupies all bandwidth and is transmitted to the target computing node; after the task with priority I is transmitted, three lower priority tasks are transmitted in parallel.

[0016] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, allocating computing resources to each computing node includes:

[0017] Distribute part of the computing resources of the cloud server to all fog servers within the range. The total computing resources that can be allocated to the fog servers are:

[0018]

[0019] Where θ1 and θ2 are the proportions of the distributed computing resources of the cloud server and fog server to their original computing resources, respectively. C The computing resources of the cloud server. is the computing resource of each CPU of the fog server, H is the number of fog nodes, CN i The number of CPUs for each fog node.

[0020] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, determining the vehicle task unloading mode according to the task information includes:

[0021] When the task vehicle occupies all bandwidth to transmit all tasks to the fog node before leaving the fog area, the unloading mode of the task vehicle is the cloud fog vehicle unloading mode, otherwise it is the service vehicle unloading mode;

[0022] Then, the unloadable service vehicles of the above two unloading modes are determined. The relative speed between the task vehicle and the idle vehicle obeys the normal distribution. When the probability of the maximum relative speed is greater than a given threshold, the idle vehicle is used as a service vehicle.

[0023] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, the delay model is a delay model of multi-vehicle multi-task parallel computing, specifically including:

[0024] ① When the task of the task vehicle is unloaded locally, the tasks are executed in sequence, and the completion time of local unloading is the sum of the computational delays of all tasks of all vehicles processed locally, T VT ;

[0025] ② When the task of the task vehicle is not unloaded locally, the task transmission between multiple task vehicles is a parallel process; the task transmission time of priority I is expressed as:

[0026]

[0027] in, is the set of tasks performed by the service vehicle, is the set of tasks executed by the fog nodes, U C It is a set of tasks executed on the cloud server. j,k It is a mission vehicle VT j mission; is the time delay for the task vehicle to transmit the task to the service vehicle, is the time delay for the task vehicle to transmit the task to the current fog node, is the delay of task transmission from current fog node to target fog node, is the time delay for the task to be transmitted from the current fog node to the cloud server; the time it takes for the task vehicle to complete the transmission of all tasks with priority I is expressed as:

[0028]

[0029] Among them, r j,k For task j,k Priority;

[0030] When a task with priority I arrives at a computing node, it is necessary to consider whether the computing node has unfinished tasks and update the completion time of the computing node using the following formula:

[0031]

[0032] in, is the computational latency of the task processing in the service vehicle, is the computational latency of the task on any CPU of the fog node, is the computing delay of the task on the cloud server; when all the tasks with priority I are completed, the completion time of each computing node for the task with priority I is expressed as Where e1 represents the last task with priority I on each computing node;

[0033] The transmission time of any task vehicle with a transmission priority of II, III, or IV to the service vehicle, fog node, and cloud server is expressed as:

[0034]

[0035] Among them, r is the priority of the task, VT for mission vehicles j The completion time of the previous task with the same priority being transferred to the target computing node;

[0036] The calculation completion time of each computing node is expressed as:

[0037]

[0038] Then the final completion time of all computing nodes is Where e2 represents the last arriving task on the computing node, so the final completion time of the task unloaded by the service vehicle is The final completion time of the task offloaded at the fog node is The final completion time of the task offloaded to the cloud server is Then the final completion time of multiple task vehicles is expressed as:

[0039] T=max(T VT ,T VI ,T F ,T C ).

[0040] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, the energy consumption model is:

[0041] When the task is unloaded locally, the energy consumption generated is the total energy consumption of the task vehicle unloading locally;

[0042] When the task is unloaded at the service vehicle, the energy consumption generated comes from the energy consumption generated by the task transmission to the service vehicle and the energy consumption generated by the task calculation at the service vehicle;

[0043] When a task is executed at a fog node, the energy consumption generated comes from the energy consumed by the task vehicle transmitting the task to the current fog node, then from the current fog node to the target fog node, and the energy consumed by the task calculation at the target fog node.

[0044] When the task is executed on the cloud server, the energy consumption generated comes from the energy consumption generated by the task vehicle transmitting the task to the current fog node, and then from the current fog node to the cloud server, and the energy consumption generated by the task calculation on the cloud server.

[0045] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, with delay and energy consumption as optimization objectives, the task unloading problem is regarded as a multi-objective optimization problem; a multi-objective optimization algorithm based on decomposition is proposed, specifically: first, the population is initialized by an initialization strategy based on reverse learning; the fitness value is obtained by the aggregation function after problem decomposition and standardization; then, the diversity of solutions is enhanced by introducing global learning, and local mutation is used to help the algorithm jump out of the local optimum, and finally an out-of-bounds processing method based on symmetry is designed to repair the out-of-bounds solution set.

[0046] According to the multi-vehicle priority task unloading resource allocation method based on IMOEA / D of the present invention, further, the cross-boundary processing method based on the symmetric idea is as follows:

[0047]

[0048] Among them, ct j Indicates the midpoint of the vehicle coding range for this column, mod(X x,y ,2ct j ) is to perform a modulo operation on the out-of-bounds coded value, X′x,y Represents the encoded value after out-of-bounds processing, and rand([-1,1]) represents a randomly generated number in {-1,0,1}.

[0049] Furthermore, the present invention also proposes a multi-vehicle priority task unloading resource allocation system based on IMOEA / D, comprising:

[0050] The multi-vehicle priority task unloading model building module is used to divide the tasks of all task vehicles into four priorities, allocate corresponding communication resources to the tasks of different priorities of the vehicles and allocate computing resources to each computing node; determine the vehicle task unloading mode according to the task information;

[0051] The optimization target building module is used to establish the delay model and energy consumption model of tasks with different priorities during the task transmission and task execution stages;

[0052] The solution module is used to design a decomposition-based multi-objective optimization algorithm to solve the optimal task offloading decision.

[0053] The beneficial effects achieved by adopting the above technical solution are:

[0054] The present invention proposes a multi-vehicle priority task unloading resource allocation method based on IMOEA / D, which can obtain the optimal task unloading strategy for multiple vehicles; ① Design a multi-vehicle task priority division and resource allocation scheme, which divides the multi-vehicle tasks into four priorities according to information such as task attributes and computing resources in the environment, and allocates corresponding communication resources for the four priority tasks. The vehicle quickly transmits high-priority tasks to the destination node according to the allocated communication bandwidth to ensure that the high-priority tasks can be processed in time; at the same time, tasks with lower priorities can be transmitted in parallel, which reduces the idle time of computing nodes while ensuring the efficiency of task completion. ② Establish a delay model and energy consumption model for tasks of different priorities in the task transmission and task execution stages, wherein the delay model considers the parallel computing relationship between each computing node, and multiple tasks can be executed simultaneously on different computing nodes. The model fully considers the parallel relationship between each computing node in the system and can reflect the task unloading process of real scenes. ③ A multi-objective optimization algorithm (IMOEA / D) based on decomposition is proposed, which innovatively proposes an out-of-bounds processing method based on symmetry. This method can solve the problem that the current out-of-bounds coding processing scheme is difficult to use the chromosome information obtained by this iteration and is prone to fall into local optimality. The IMOEA / D algorithm can obtain solutions with lower completion time and energy consumption in the proposed multi-vehicle priority task unloading model. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings of the embodiments of the present invention, wherein the drawings are only used to illustrate some embodiments of the present invention, but not to limit all embodiments of the present invention thereto.

[0056] Figure 1 is a model diagram of a multi-vehicle task unloading system according to an embodiment of the present invention;

[0057] Figure 2 is a bandwidth allocation diagram for tasks of different priorities according to an embodiment of the present invention;

[0058] Figure 3 is a road condition analysis diagram of an embodiment of the present invention;

[0059] Figure 4 is a delay model of parallel computing in an embodiment of the present invention;

[0060] Figure 5 is an example of multi-vehicle task encoding according to an embodiment of the present invention;

[0061] Figure 6 is an example of an initialization strategy based on reverse learning according to an embodiment of the present invention;

[0062] Figure 7 This is an example of population crossover and mutation operation in an embodiment of the present invention;

[0063] Figure 8 This is an example of a cross-border processing method based on the symmetry concept in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings of specific embodiments of the present invention to clearly and completely describe the exemplary scheme of the embodiment of the present invention. Unless otherwise defined, the technical terms or scientific terms used in the present invention should be the common meanings understood by people with ordinary skills in the field.

[0065] This embodiment discloses a multi-vehicle priority task unloading resource allocation method based on IMOEA / D, comprising the following steps:

[0066] Step S101: Divide the tasks of all task vehicles into four priorities, and allocate corresponding communication resources to the tasks of vehicles with different priorities and allocate computing resources to each computing node.

[0067] Step S102: determining the vehicle task unloading mode according to the task information, and establishing a delay model and an energy consumption model for tasks of different priorities during the task transmission and task execution stages.

[0068] Step S103: design a decomposition-based multi-objective optimization algorithm to solve the optimal task offloading decision.

[0069] (1) System model

[0070] The multi-vehicle task offloading system model in the vehicle fog computing scenario is as follows Figure 1 As shown in Figure 1, multiple roadside units (RSUs) are evenly distributed on both sides of a two-way road. Each RSU is equipped with a computing server to form a fog node. The service range of the fog node is called the fog area. Assume that there are H fog nodes in the system, denoted by F = {F1, F2, …, F i ,…F H}, where 1≤i≤H. At a certain moment, there are N vehicles in the considered scenario with task offloading requirements, denoted as VT={VT1,VT2,…,VT j ,…VT N}, where 1≤j≤N. The vehicle without task unloading demand is defined as an idle vehicle, and the idle vehicle that establishes a connection with the task vehicle and provides unloading service is defined as a service vehicle. The idle vehicle is recorded as VI = {VI1, VI2, ..., VI q ,…VI M}, where 1≤q≤M. The tasks to be processed for each task vehicle can be divided into S j subtasks, and the subtask set corresponding to all task vehicles is recorded as U = {U1, U2, …, U j ,…U N}, where 1≤j≤N. The subtask set of each task vehicle is represented as Where 1≤k≤S j Table 1 summarizes the main symbols used in this scheme.

[0071] Table 1 Description of main symbols

[0072]

[0073]

[0074] (2) Vehicle distribution model

[0075] In order to build a more realistic road environment model, the number of vehicles, vehicle spacing and vehicle speed changes in the road section are represented in the form of Poisson distribution, exponential distribution and normal distribution respectively.

[0076] The arrival interval distance of vehicles entering a certain road section follows the Poisson distribution. The number of vehicles within the coverage area of ​​all fog nodes is The probability can be expressed as:

[0077]

[0078] Among them, δ1 is the traffic density parameter of the road section. Finally, N vehicles are randomly set as task vehicles and the remaining vehicles are set as idle vehicles.

[0079] The distance between vehicles follows an exponential distribution, and the probability that the distance between two vehicles is l can be expressed as:

[0080]

[0081] Among them, δ2 is the vehicle interval parameter of the road section, and l≥10 is to ensure the safe distance between vehicles.

[0082] Assuming that the speed change in the road section follows a normal distribution, when the fog node receives the task information, the speed v j The probability density function of can be estimated as:

[0083]

[0084] The estimated instantaneous speed of the vehicle As the vehicle's true speed.

[0085] (3) Task Prioritization and Resource Allocation Model

[0086] In order to make fuller use of environmental resources and ensure the timely completion of tasks, all tasks of task vehicles are divided into four priorities, and corresponding communication resources are allocated to tasks of different priorities. The task priority value is determined by three indicators: the deadline of the task, the amount of data of the task, and the amount of calculation of the task. For tasks with shorter deadlines, they should have higher priorities. According to queuing theory, tasks with smaller data volumes and calculation volumes should have higher priorities. Therefore, the priority value of the task can be obtained by the following formula:

[0087] Pri k =α·τ j,k +β·ω j,k +γ·c j,k

[0088] Among them, Pri k is the task priority value. The smaller the value, the higher the priority level of the task. α, β, and γ are the weight factors of the priority values ​​corresponding to the task deadline, the amount of data of the task, and the amount of calculation of the task, respectively. After obtaining the priority values ​​of all tasks, all tasks are arranged in ascending order according to the priority values, and the tasks are divided into four priority levels. If they cannot be evenly divided, the tasks are prioritized as high priority tasks.

[0089] Regarding the allocation of communication resources, the bandwidth allocation of tasks with different priorities of multiple vehicles is considered. First, in order to improve the reliability of communication between task vehicles and fog nodes, more bandwidth needs to be allocated to vehicles closer to the edge of the fog area. The bandwidth share of task vehicles can be expressed as:

[0090]

[0091] Furthermore, the bandwidth allocated when the task vehicle communicates with the fog node can be expressed as:

[0092]

[0093] Each service vehicle can only be connected to one task vehicle, so the link between the task vehicle and the idle vehicle does not require bandwidth allocation. The remaining links are all wired links and do not require bandwidth allocation.

[0094] Next, the bandwidth allocated to the task vehicle is redistributed to allocate corresponding bandwidth to tasks of different priorities in each task vehicle. The specific allocation is as follows: Figure 2 As shown in the figure. When the task starts to be transmitted, the task with priority I on the task vehicle occupies all bandwidth to be transmitted to the target computing node. After the task with priority I is transmitted, the three lower priority tasks occupy and The tasks are transmitted in parallel with the link bandwidth of , and B is the bandwidth set of the task transmission link.

[0095] Regarding the allocation of computing resources, considering that the cloud server can provide a large amount of computing resources for the vehicle, due to the long distance between the cloud server and the vehicle, the vehicle will have a long delay when communicating with the cloud server. Therefore, part of the computing resources of the cloud server can be distributed to the fog servers around the vehicle in advance through service migration, so as to enhance the computing power of the fog server. After the fog server receives the computing resources from the cloud server, it integrates and allocates the computing resources of all fog servers within the range to ensure the timely completion of the task. The total computing resources that the fog server can allocate are:

[0096]

[0097] Among them, θ1 and θ2 are the proportion of the distributed computing resources of the cloud server and fog server to their original computing resources. The computing resources will be allocated between fog servers according to the number of task vehicles within their service range. The more task vehicles there are, the more computing resources will be allocated; conversely, the fewer computing resources will be allocated. Multiple CPUs of each fog server evenly distribute computing resources. The computing resources of each CPU of the fog server can be expressed as:

[0098]

[0099] Among them, VN i represents the fog server f i F The number of mission vehicles within range.

[0100] (4) Vehicle task offloading mode

[0101] Due to the mobility of the vehicle, the communication link between the vehicle and the surrounding communication equipment will change as the vehicle moves, and the reliability of task transmission cannot be guaranteed. This article divides the offloading modes into two types, namely, the cloud fog vehicle offloading mode and the service vehicle offloading mode. The cloud fog vehicle offloading mode can offload tasks to the cloud server, fog server, service vehicle and local; the service vehicle offloading mode can only offload tasks to the service vehicle and local. The basis for determining the two offloading modes is: when the task vehicle can occupy all bandwidth to transmit all tasks to the fog node before leaving the fog area, the offloading mode of the task vehicle is the cloud fog vehicle offloading mode, otherwise it is the service vehicle offloading mode. The delay of the task vehicle occupying all bandwidth to transmit all tasks can be expressed as:

[0102]

[0103] Among them, B v2f is the communication bandwidth between the vehicle and the fog node. The time it takes for the task vehicle to leave the current fog area is:

[0104]

[0105] in, is the distance between the vehicle and the edge of the fog area.

[0106] Both the cloud vehicle unloading mode and the service vehicle unloading mode can unload tasks to surrounding vehicles, and it is necessary to determine the unloading service vehicles for the two unloading modes. Since the speed of the vehicle is changing, it is difficult to predict the precise communication time between vehicles. Therefore, when the communication time between the task vehicle and the idle vehicle is greater than the time for the two vehicles to transmit all tasks, the probability is greater than a given threshold ξ, then the idle vehicle is considered to be a service vehicle.

[0107] Since the speed of a single vehicle follows a normal distribution, the relative speed between the task vehicle and the idle vehicle also follows a normal distribution:

[0108]

[0109] Where Δv represents the average relative speed between the task vehicle and the idle vehicle, is the difference between the average speed of the task vehicle and the idle vehicle, and Δs is the difference between the standard deviation of the speed of the task vehicle and the idle vehicle. The communication time between vehicles is determined by the relative speed and location of the vehicles. The critical speed at which the two vehicles can maintain connection before the communication ends is called the maximum relative speed of the two vehicles. The speed and location of the task vehicle and an idle vehicle are as follows: Figure 3 As shown, suppose the distance between vehicles in each case is d v2v ,but Figure 3 The maximum relative speed of the two vehicles in the four cases on the left is:

[0110]

[0111] The maximum relative speeds of the two vehicles in the four cases on the right are:

[0112]

[0113] In the above two formulas, is the maximum communication distance of V2V, is the time it takes for the task vehicle to transfer all tasks to the idle vehicle. max )>ξ, the idle vehicle is considered to have high reliability and can be used as a service vehicle.

[0114] (5) Delay model

[0115] When the mission vehicle VT j task j,k When offloading locally, the task does not need to be transmitted, and the computational latency of the task processed locally can be expressed as:

[0116]

[0117] When the mission vehicle VT j task j,k In Service Vehicle VI q During execution, the task vehicle needs to transfer the task to the service vehicle. The path loss between the task vehicle and the service vehicle is expressed as The channel gain between the task vehicle and the idle vehicle is Then the transmission rate between the task vehicle and the service vehicle is:

[0118]

[0119] in, It is a task j,k Given the transmission bandwidth between vehicles, the delay for the task vehicle to transmit the task to the service vehicle is:

[0120]

[0121] The computational delay of the task in the service vehicle is:

[0122]

[0123] The amount of processing result data in the task return phase is usually small, so the task return delay is ignored.

[0124] When the mission vehicle VT j task j,kWhen executing at a fog node, the task vehicle needs to determine whether the task needs to be transmitted across fog nodes. When the task is executed at the current fog node, there is no need to consider the transmission of tasks across fog nodes. The path loss between the task vehicle and the current fog node is in is the distance between the vehicle and the current fog node. The signal gain between the task vehicle and the fog node is Then the transmission rate between the task vehicle and the current fog node is:

[0125]

[0126] in, It is a task j,k The transmission bandwidth between the task vehicle and the fog node is:

[0127]

[0128] When the task is not executed on the current fog node, the task needs to be transmitted to the target fog node through the fog node in a multi-hop manner. The transmission rate of the task between fog nodes is:

[0129]

[0130] in, It is a task j,k Given the transmission bandwidth between fog nodes, the transmission delay of the task from the current fog node to the target fog node can be expressed as:

[0131]

[0132] Among them, h f2f is the number of hops from the current fog node to the target fog node.

[0133] The computational latency of a task on any CPU of a fog node is:

[0134]

[0135] When the mission vehicle VT j task j,k When the cloud server is executing, the task vehicle first transmits the task to the current fog node, and then transmits it to the cloud server from the fog node. The transmission rate between the fog node and the cloud server is:

[0136]

[0137] in, It is a task j,kThe transmission bandwidth between the fog node and the cloud server is , then the transmission delay from the current fog node to the cloud server can be expressed as:

[0138]

[0139] The direct computing latency of the task on the cloud server is:

[0140]

[0141] The tasks of all vehicles are transmitted to the target location in parallel for execution. The tasks of different unloading locations are parallel calculation processes, and the tasks of different priorities of the same vehicle are also parallel calculation processes. Therefore, the final completion time of all task vehicles is not the direct sum of the delays of each part. In order to describe this unloading process and adopt the above resource allocation scheme, this scheme constructs a delay model of multi-vehicle multi-task parallel computing.

[0142] ① When the tasks of the task vehicle are unloaded locally, the tasks are executed sequentially. Therefore, the completion time of local unloading can be expressed as:

[0143]

[0144] in, A collection of tasks that are executed locally.

[0145] ② When the task of the task vehicle is not unloaded locally, the task transmission between multiple task vehicles is a parallel process. At the beginning of unloading, the task with priority I occupies all bandwidth to be transmitted to the target computing node, and then each task vehicle transmits the tasks of subsequent priorities in parallel. The transmission time of the task with priority I can be expressed as:

[0146]

[0147] in, is the set of tasks performed by the service vehicle, is the set of tasks executed by the fog nodes, U C is the set of tasks executed on the cloud server. The time it takes for the task vehicle to complete the transmission of all priority I tasks is expressed as:

[0148]

[0149] The task transmission process between task vehicles is in parallel mode, and tasks of different vehicles will occupy computing resources according to the order in which the tasks arrive, that is, the tasks are processed in a "first come, first served" mode. Therefore, when a task with priority I arrives at a computing node, it is necessary to consider whether the computing node has unfinished tasks, and use the following formula to update the completion time of the computing node:

[0150]

[0151] When all the tasks with priority I are calculated, the completion time of each computing node for the task with priority I can be expressed as Where e1 represents the last task with priority I on each computing node.

[0152] Tasks with different priorities of II, III and IV are transmitted in parallel, and tasks with the same priority are transmitted serially. Tasks with different priorities can be transmitted in parallel to the same target computing node. After reaching the target computing node, the task needs to seize the computing resources of the computing node, and the transmission completion time of each task to the computing node needs to be determined. The transmission time of any task vehicle with a transmission priority of II, III, IV to the service vehicle, fog node and cloud server is expressed as:

[0153]

[0154] Among them, r is the priority of the task, VT for mission vehicles j The completion time of the previous task with the same priority in the transmission to the target computing node; when k-1=0, it indicates the start time of the transmission of tasks with priorities II, III, and IV, and the completion time of the transmission of tasks with priority I

[0155] When a task is transferred to the target computing node and the computing node is in an idle state, the task is executed in the order in which it arrives. If the current computing node has unfinished tasks, the tasks that have arrived will be executed in sequence after the tasks are completed. The computing completion time of each computing node is expressed as:

[0156]

[0157] Then the final completion time of all computing nodes is Where e2 represents the last arriving task on the computing node. Therefore, the final completion time of the task unloaded by the service vehicle is The final completion time of the task offloaded at the fog node is The final completion time of the task offloaded to the cloud server is Then the final completion time of multiple task vehicles can be expressed as:

[0158] T=max(T VT ,T VI ,T F ,T c )

[0159] In order to better explain the delay model of parallel computing, Figure 4 The unloading process of a task vehicle in the environment is used as an example. Assume that the task vehicle is connected to a service vehicle, and there are two fog nodes around the task vehicle that provide unloading services, among which the number of CPUs of fog node 1 is 2 and the number of CPUs of fog node 2 is 1. The task vehicle divides the tasks to be processed into 22 subtasks, so the number of tasks with priorities I, II, III and IV are 6, 6, 5, and 5 respectively. The tasks unloaded locally start execution directly. At the same time, the tasks with priority I occupy all bandwidth to start transmission. When they are transmitted to the target computing node, it is determined whether the current node is idle. When the computing node has unfinished tasks, it starts execution after the task is completed, such as task1 in the figure. When the last task task2 with priority I is transmitted, the subsequent priority tasks are immediately transmitted. Since fog node 1 has two CPUs and one CPU is idle, task3 can be executed on fog node 1 at the same time as another task. The task task4 unloaded by the service vehicle needs to be executed after the task that arrived before task4 is completed. When all tasks of each computing node are completed, the end time of the last completed task, task5, is the final completion time of the task vehicle.

[0160] (6) Energy consumption model

[0161] When the task is offloaded locally, the energy consumption mainly comes from the task calculation process, and the energy consumption of this process is:

[0162]

[0163] The total energy consumption of the task vehicle for local unloading is

[0164] When the task is unloaded at the service vehicle, the energy consumption generated comes from the task transmission process and the task calculation process. The energy consumption generated by the task transmission process to the service vehicle is:

[0165]

[0166] The energy consumption generated by the task in the service vehicle calculation process is:

[0167]

[0168] The total energy consumption of the mission vehicle during the service vehicle operation is:

[0169]

[0170] in, is the set of tasks performed by the service vehicle, It is a mission vehicle VT j A collection of unloadable service vehicles.

[0171] When the task is executed at the fog node, the energy consumption generated comes from the transmission process of the task vehicle transmitting the task to the current fog node, and then from the current fog node to the target fog node, and the calculation process of unloading the task at the target fog node. The energy consumption generated by the task vehicle transmitting the task to the current fog node is:

[0172]

[0173] The energy consumption of the current fog node forwarding the task to the target fog node is:

[0174]

[0175] The energy consumption generated by the task calculation at the target fog node is:

[0176]

[0177] The total energy consumption of the task vehicle executed in the fog node is:

[0178]

[0179] in, It is a set of tasks executed on the fog nodes.

[0180] When the task is executed on the cloud server, the energy consumption generated comes from the task vehicle transmitting the task to the current fog node, and then from the current fog node to the cloud server, and the task calculation process on the cloud server. The energy consumption generated by the task being transmitted from the current fog node to the cloud server is:

[0181]

[0182] The energy consumption of the task in the cloud server is:

[0183]

[0184] The total energy consumption of the mission vehicle executed on the cloud server is:

[0185]

[0186] in, It is a collection of tasks executed on the cloud server.

[0187] The total energy consumption of the task offloading process of all vehicles can be expressed as:

[0188]

[0189] (7) Problem Model

[0190] This scheme studies the problem of unloading tasks of different priorities from multiple vehicles in the cloud-fog collaborative Internet of Vehicles scenario. After allocating computing resources, the multi-vehicle priority task unloading problem is regarded as a constrained multi-objective optimization problem with latency and energy consumption as optimization objectives, and the optimization problem is expressed as:

[0191] minT,E

[0192]

[0193] r j,k ∈{1,2,3,4},1≤j≤N,1≤k≤S j (4)

[0194] Among them, formula (1) indicates that the final completion time of all vehicles’ tasks cannot exceed the sum of the average deadline delays of each vehicle: Formula (2) and (3) indicate that each vehicle's task can only be executed at one location, and each subtask must be executed; Formula (4) indicates that the task priority is divided into four levels, and each task has a corresponding priority value.

[0195] (8) Decomposition-based multi-objective optimization algorithm (IMOEA / D)

[0196] The main idea of ​​MOEA / D is to decompose the multi-objective optimization problem into multiple sub-problems using a set of uniformly distributed normalized weight vectors. Each sub-problem obtains information by using several of its neighboring sub-problems and continuously improves the solution of the current sub-problem. On this basis, this scheme proposes a new improved decomposition-based multi-objective optimization algorithm (IMOEA / D), which makes improvements under the framework of the original MOEA / D algorithm. Specifically, a new encoding method suitable for multi-vehicle task offloading is designed; the population is initialized in a reverse learning-based manner; and a global learning and local perturbation balance method is introduced.

[0197] ①Multi-vehicle task encoding based on priority block

[0198] Encode all tasks of different priorities for multiple vehicles at the same time, and divide the chromosomes into numbered chromosomes representing task information and common chromosomes representing task unloading locations. There are two numbered chromosomes, namely the vehicle number chromosome and the task number chromosome. The two chromosomes are used together to determine the source of each column of tasks. For common chromosomes, set the encoding value of each bit to Among them, Pn represents the population size, and Tn represents the total number of tasks of all task vehicles. x,y =0, indicating that the task is offloaded locally; when 1≤X x,y When ≤H, it means that the task is unloaded at the fog node with the corresponding number; when Xx,y =H+1 indicates that the task is offloaded to the cloud server; when X x,y ≥H+1, it means that the task number is X x,y -(H+1) idle vehicles are unloaded. On this basis, each chromosome is divided into four equal-length sub-chromosomes, each of which represents the unloading location of all tasks of a certain priority. If the chromosome cannot be divided into four equal parts, the task with a higher priority is set first.

[0199] Figure 5 An example of encoding a set of chromosomes is shown. Assume that there are 3 task vehicles in the environment, among which task vehicle No. 1 divides the task into 5 subtasks and is not connected to the unloading vehicle. Task vehicle No. 2 divides the task into 4 subtasks and is connected to 2 unloading vehicles. Task vehicle No. 3 divides the task into 6 subtasks and is connected to 1 unloading vehicle. First, based on the task information of all task vehicles, the priority value of each task is calculated. Then, the priority level is assigned after sorting according to the priority value. Since all tasks cannot be evenly divided into four priorities, tasks with higher priorities are given priority. Figure 5 In , the first chromosome represents the vehicle number, the second chromosome represents the task number of the task vehicle corresponding to the first chromosome, and the remaining chromosomes represent the unloading location of a task corresponding to a task vehicle. For example, X 1,2 =0 means the 5th task of the first task vehicle is unloaded locally; X 2,5 =6 means the third task of the second task vehicle is unloaded at the idle vehicle numbered 2. Pn,12 =2 means that the fifth task of the third vehicle is unloaded at the fog node numbered 2.

[0200] ② Population initialization based on reverse learning

[0201] In the population initialization stage, reverse learning is introduced to obtain an initial population with higher diversity. First, half of the population is randomly initialized, and then reverse learning is performed on the initialized general population to obtain the other half of the population. The randomly initialized population and the reverse learning population are merged to obtain the final initial population. The reverse learning process can be expressed as:

[0202]

[0203] Among them, ct j Indicates the midpoint of the vehicle coding range for this column of tasks.

[0204] Figure 6 So Figure 5The reverse learning example of the first common chromosome in . The first bit of the code represents the unloading location of the second task of the task vehicle No. 3, and the task vehicle No. 3 is connected to an unloading vehicle. From the above coding rules, it can be seen that its coding range is X x,y ∈[0,5], the midpoint of the coding range is ct3=2.5. According to the above formula, when the first bit coding value is 1, the coding value after reverse learning is 4. Similarly, the 7th bit coding and the 11th bit coding represent the unloading position of the 3rd task of the No. 1 task vehicle and the unloading position of the 4th task of the No. 2 task vehicle, respectively. The coding range of the No. 1 task vehicle is X x,y ∈[0,4], the midpoint of the coding range is ct1=2; the coding range of vehicle No. 2 is X x,y ∈[0,6], the midpoint of the coding range is ct2 = 3. The coding values ​​after reverse learning calculated from the above formula are 0 and 6 respectively.

[0205] ③Problem decomposition and aggregation functions

[0206] MOEA / D decomposes the multi-objective problem into multiple sub-problems for optimization based on the weight vector. Each sub-problem corresponds to a weight vector. The uniformity of the weight vector will have a certain impact on the performance of the algorithm. The solution set obtained by a more uniform weight vector will be better. Therefore, IMOEA / D uses the simplex lattice method to generate a uniform weight vector. The formula of this method is:

[0207]

[0208] Among them, λ w is the weight vector, Q is the number of weight vectors, and must satisfy D is the number of multi-objective optimization problems.

[0209] IMOEA / D uses the Chebyshev method to aggregate sub-problems into a single-objective optimization problem. This method approaches the Pareto solution by reducing the maximum distance between the individual and the ideal point, and is applicable to non-continuous optimization problems. The Chebyshev method can be expressed as:

[0210]

[0211] Among them, z * is the ideal point, f d is the function value of the subproblem in dimension d, is the optimal value of the subproblem in dimension d.

[0212] Due to different dimensions, there is a large difference in the values ​​of delay and energy consumption. In order to avoid the algorithm focusing on optimizing the larger numerical targets and ignoring the smaller numerical issues, it is necessary to standardize the optimization targets. The standardized formula used by IMOEA / D can be expressed as:

[0213]

[0214] in, is the maximum value of the subproblem in dimension d, is the minimum value of the subproblem in dimension d, and ψ is a small constant.

[0215] ④ Population crossover and mutation

[0216] IMOEA / D uses single-point crossover and random mutation to operate chromosomes. The chromosome interaction process in the population only operates on the common chromosomes in the population, and the numbered chromosomes do not participate in the operation process. Therefore, the crossover and mutation operations are mainly aimed at common chromosomes.

[0217] Single-point crossover randomly selects two chromosomes and randomly selects a crossover point, and exchanges the gene fragments of the two chromosomes from the crossover point to obtain two different chromosomes. After the single-point crossover operation is completed, the two new chromosomes are subjected to random mutation operation. First, each bit code of the new chromosome is traversed and a random number is generated bit by bit. When the random number corresponding to a certain bit code is less than the given mutation probability, the value of the bit code is randomly set to any value within the code range.

[0218] Figure 7 The process of crossover mutation is shown. First, chromosome 1 and chromosome 2 are randomly selected and the 10th position is randomly set as the crossover point. After chromosome 1 and chromosome 2 exchange 10 positions, the gene fragments are obtained to obtain a new chromosome. Then, the 4th and 10th positions of the new chromosome 1 and the 8th position of the new chromosome 2 are changed in the coding range to complete the mutation operation. Finally, two new chromosomes are obtained.

[0219] ⑤Global learning and local perturbations

[0220] The MOEA / D algorithm iteratively optimizes the weight vector through information interaction between neighborhood sub-problems. During the operation of the algorithm, the solutions of adjacent weight vectors are relatively similar, and it is impossible to obtain global chromosome information, making it difficult to obtain a better solution and easily falling into the local optimum. Therefore, the IMOEA / D algorithm introduces a global learning strategy and a random difference mutation strategy to improve population diversity, speed up the convergence of the algorithm, and escape from the local optimum.

[0221] The global learning strategy means that during the algorithm iteration process, chromosomes are no longer limited to obtaining chromosome information between neighbors. By increasing the opportunities for mutual learning between populations, the diversity of the population is improved and the convergence speed of the algorithm is accelerated. The global learning method in the population can be expressed as:

[0222] X x ′ ,y =X x,y +round(rand×|X x,y -X x′ , y |)

[0223] Among them, X x ' ,y represents the chromosome after global learning, round() represents rounding to the nearest integer, rand is a random number between 0 and 1, and X x′,y represents a chromosome randomly selected from the population.

[0224] Local perturbation uses a random difference mutation strategy, which helps the algorithm escape from the local optimum by obtaining information from the optimal solution and the random solution. The random difference mutation strategy is expressed as:

[0225]

[0226] in, is the global optimal solution. In order to reduce the complexity of the algorithm and speed up the convergence of the algorithm, the random difference mutation strategy is only used for some random individuals when the optimal solution of the algorithm has not changed for θ consecutive generations.

[0227] ⑥ Cross-border processing method based on symmetry

[0228] After global learning and local perturbations, some values ​​of chromosome encoding will exceed the corresponding encoding range, so it is necessary to process the out-of-bounds encoding. At present, most of the solutions for out-of-bounds encoding are boundary assignment method or penalty function method. The former directly assigns the out-of-bounds encoding value to the boundary value of the out-of-bounds direction, while the penalty function method penalizes the fitness value of the chromosome that exceeds the boundary to reduce the probability of out-of-bounds encoding in subsequent iterations. Although these solutions can handle out-of-bounds encoding, it is difficult for these solutions to utilize the chromosome information obtained in this iteration and they are prone to fall into local optimality. In the IMOEA / D algorithm, a coding out-of-bounds processing solution based on symmetry is proposed. This solution retains the chromosome information obtained from adjacent subproblems, adds random factors, and ensures the diversity of the population. The specific methods for out-of-bounds processing are as follows:

[0229]

[0230] In the formula, mod(X x,y ,2ct j ) is to perform a modulo operation on the out-of-bounds encoding value. This operation is to ensure that the value that exceeds multiple encoding ranges can fall into the encoding range after an out-of-bounds processing. x,yRepresents the encoded value after out-of-bounds processing, and rand([-1,1]) represents a randomly generated number in {-1,0,1}.

[0231] Taking the coding example scenario as an example, Figure 8 The process of out-of-bounds handling is shown in Figure 1. (a) The figure shows the process of out-of-bounds handling of the code of the vehicle in task 2. x,y >2ct j , with the nearest boundary point 2ct j is the symmetry axis, and the symmetric point in the coding range is 3, so the processed coding value will be selected from 2, 3, and 4. Similarly, Figure (b) shows the coding out-of-bounds processing process of the No. 3 task of the No. 1 task vehicle. In the figure, X x,y <0, the nearest boundary point 0 is taken as the symmetry axis, and the symmetry point within the encoding range is 2, so the processed encoding value will be selected from 1, 2, and 3.

[0232] Corresponding to the above method, this embodiment also discloses a multi-vehicle priority task unloading resource allocation system based on IMOEA / D, including:

[0233] The multi-vehicle priority task offloading model construction module is used to divide the tasks of all task vehicles into four priorities, allocate corresponding communication resources to the tasks of different priorities of the vehicles and allocate computing resources to each computing node; determine the vehicle task offloading mode according to the task information.

[0234] The optimization target building module is used to establish the delay model and energy consumption model of tasks with different priorities in the task transmission and task execution stages.

[0235] The solution module is used to design a decomposition-based multi-objective optimization algorithm to solve the optimal task offloading decision.

[0236] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of combination of hardware and software.

[0237] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A multi-vehicle priority task unloading resource allocation method based on IMOEA / D, characterized in that: Include: Divide the tasks of all mission vehicles into four priorities, and allocate corresponding communication resources to the tasks of different priorities of the vehicles and computing resources to each computing node; The vehicle task unloading mode is determined according to the task information, and the delay model and energy consumption model of tasks with different priorities in the task transmission and task execution stages are established; A decomposition-based multi-objective optimization algorithm is designed to solve the optimal task offloading decision.

2. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: The priority value of a task is determined by three indicators: the deadline of the task, the amount of data of the task, and the amount of calculation of the task. The smaller the priority value, the higher the priority level of the task. After obtaining the priority values ​​of all tasks, all tasks are arranged in ascending order according to the priority values, and the tasks are divided into four priorities.

3. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: The allocation of corresponding communication resources for tasks of different priorities of the vehicle includes: First, the bandwidth allocated when the task vehicle communicates with the fog node is expressed as: In the formula, Br j v2f is the bandwidth ratio of the mission vehicle, N represents the number of mission vehicles, B v2f is the communication bandwidth between vehicles and fog nodes; Next, the bandwidth allocated to the task vehicle is reallocated; the task with priority I on the task vehicle occupies all bandwidth and is transmitted to the target computing node; after the task with priority I is transmitted, three lower priority tasks are transmitted in parallel.

4. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: Allocating computing resources to each computing node includes: Distribute part of the computing resources of the cloud server to all fog servers within the range. The total computing resources that can be allocated to the fog servers are: Where θ1 and θ2 are the proportions of the distributed computing resources of the cloud server and fog server to their original computing resources, respectively. C is the computing resource of the cloud server. is the computing resource of each CPU of the fog server, H is the number of fog nodes, CN i The number of CPUs for each fog node.

5. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: Determining the vehicle task unloading mode based on task information includes: When the task vehicle occupies all bandwidth to transmit all tasks to the fog node before leaving the fog area, the unloading mode of the task vehicle is the cloud fog vehicle unloading mode, otherwise it is the service vehicle unloading mode; Then, the unloadable service vehicles of the above two unloading modes are determined. The relative speed between the task vehicle and the idle vehicle obeys the normal distribution. When the probability of the maximum relative speed is greater than a given threshold, the idle vehicle is used as a service vehicle.

6. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: The delay model is a delay model for multi-vehicle multi-task parallel computing, specifically including: ① When the task of the task vehicle is unloaded locally, the tasks are executed in sequence, and the completion time of local unloading is the sum of the computational delays of all tasks of all vehicles processed locally, T VT ; ② When the task of the task vehicle is not unloaded locally, the task transmission between multiple task vehicles is a parallel process; the task transmission time of priority I is expressed as: in, is the set of tasks performed by the service vehicle, is the set of tasks executed by the fog nodes, U C It is a set of tasks executed on the cloud server. j,k It is a mission vehicle VT j mission; is the time delay for the task vehicle to transmit the task to the service vehicle, is the time delay for the task vehicle to transmit the task to the current fog node, is the delay of task transmission from current fog node to target fog node, is the time delay for the task to be transmitted from the current fog node to the cloud server; the time it takes for the task vehicle to complete the transmission of all tasks with priority I is expressed as: Among them, r j,k For task j,k Priority; When a task with priority I arrives at a computing node, it is necessary to consider whether the computing node has unfinished tasks and update the completion time of the computing node using the following formula: in, is the computational latency of the task processing in the service vehicle, is the computational latency of the task on any CPU of the fog node, is the computing delay of the task on the cloud server; when all the tasks with priority I are completed, the completion time of each computing node for the task with priority I is expressed as Where e1 represents the last task with priority I on each computing node; The transmission time of any task vehicle with a transmission priority of II, III, or IV to the service vehicle, fog node, and cloud server is expressed as: Among them, r is the priority of the task, VT for mission vehicles j The completion time of the previous task with the same priority being transferred to the target computing node; The calculation completion time of each computing node is expressed as: Then the final completion time of all computing nodes is Where e2 represents the last arriving task on the computing node, so the final completion time of the task unloaded by the service vehicle is The final completion time of the task offloaded at the fog node is The final completion time of the task offloaded to the cloud server is Then the final completion time of multiple task vehicles is expressed as: T=max(T VT ,T VI ,T F ,T c )。 7. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: The energy consumption model is: When the task is unloaded locally, the energy consumption generated is the total energy consumption of the task vehicle unloading locally; When the task is unloaded at the service vehicle, the energy consumption generated comes from the energy consumption generated by the task transmission to the service vehicle and the energy consumption generated by the task calculation at the service vehicle; When a task is executed at a fog node, the energy consumption generated comes from the energy consumed by the task vehicle transmitting the task to the current fog node, then from the current fog node to the target fog node, and the energy consumed by the task calculation at the target fog node. When the task is executed on the cloud server, the energy consumption generated comes from the energy consumption generated by the task vehicle transmitting the task to the current fog node, and then from the current fog node to the cloud server, and the energy consumption generated by the task calculation on the cloud server.

8. The multi-vehicle priority task unloading resource allocation method based on IMOEA / D according to claim 1 is characterized in that: With latency and energy consumption as optimization targets, the task offloading problem is regarded as a multi-objective optimization problem. A decomposition-based multi-objective optimization algorithm is proposed. Specifically, the population is initialized by an initialization strategy based on reverse learning. The fitness value is obtained through the aggregation function after problem decomposition and standardization. Then, the diversity of solutions is enhanced by introducing global learning, and local mutation is used to help the algorithm escape from the local optimum. Finally, an out-of-bounds processing method based on symmetry is designed to repair the out-of-bounds solution set.

9. The IMOEA / D-based multi-vehicle priority task unloading resource allocation method according to claim 8 is characterized in that: The specific method of handling cross-border based on symmetry is as follows: Among them, ct j Indicates the midpoint of the vehicle coding range for this column, mod(X x,y ,2ct j ) is to perform a modulo operation on the out-of-bounds coded value, X′ x,y It represents the encoded value after out-of-bounds processing, and rand([-1,1]) represents a randomly generated number in {-1,0,1}.

10. A multi-vehicle priority task unloading resource allocation system based on IMOEA / D, characterized in that: include: The multi-vehicle priority task unloading model building module is used to divide the tasks of all task vehicles into four priorities, allocate corresponding communication resources to the tasks of different priorities of the vehicles and allocate computing resources to each computing node; determine the vehicle task unloading mode according to the task information; The optimization target building module is used to establish the delay model and energy consumption model of tasks with different priorities during the task transmission and task execution stages; The solution module is used to design a decomposition-based multi-objective optimization algorithm to solve the optimal task offloading decision.

Images