Task unloading method for emergency rescue scene under space-air-ground integrated network

Abstract

The invention provides a task unloading method for an emergency rescue scene under a space-air-ground integrated network, and relates to the technical field of calculation unloading, and the method comprises the steps: unifying the feature description of task unloading; establishing a joint optimization problem model by taking minimization of unloading time delay and power consumption as an optimization target; determining a task unloading priority based on a firefly algorithm; constructing a reward mechanism of a deep reinforcement learning model according to the task unloading priority, and solving the joint optimization problem model by using a deep reinforcement learning algorithm to obtain a plurality of task unloading schemes; performing simulation verification based on the task unloading scheme to obtain evaluation data; parameters of the joint optimization problem model are corrected according to the evaluation data, and a task unloading optimal model is obtained; and solving the optimal task unloading model to obtain a final task unloading scheme. Reasonable and effective resource allocation is carried out on a large number of tasks which emerge in an emergency rescue scene and need to be processed in real time, and the scheduling efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of computing offloading technology, and in particular to a task offloading method for emergency rescue scenarios in an air-ground integrated network. Background Art

[0002] Disasters such as earthquakes, floods, and rainstorms and mudslides caused by extreme weather can cause great damage to various communication resources, hindering the rescue and reconstruction process. When natural disasters occur, a large number of urgent real-time tasks need to be processed. At present, with the full deployment of the fifth-generation communication technology 5G, many countries have begun to build the development of 6G. On the basis of 5G, 6G has built a space-air-ground integrated network (SAGIN, Space-Air-Ground Integrated Networks). Regarding the 6G network, many researchers at home and abroad have put forward their own visions. The space-air-ground integrated network can break through the limitations of the surface and realize global large-scale rapid communication services. SAGIN effectively supports seamless coverage and efficient resource pipelines. Especially for disaster-prone areas, the space-air-ground integrated network can well solve the problem of insufficient communication resources in disaster relief, so it has become one of the core research directions of the future 6G network architecture.

[0003] In recent years, many scholars at home and abroad have begun to study the application of SAGIN in task scheduling and resource management. By reviewing the evolution of mobile communication networks, they analyzed the application and technical requirements of 6G, proposed a flexible, low-latency and flat SAGIN architecture, and gave corresponding application cases. Some scholars designed whether the task was offloaded to the same server. While maximizing the throughput of the system, the task was sliced ​​so that the same task could be processed on different servers. Many scholars abstracted the task scheduling problem into a Markov decision process (MDP) and used reinforcement learning to find the allocation solution. Based on the integrated air-ground network architecture, a user satisfaction function was constructed, and the network task allocation process was abstracted into a semi-Markov decision process (SMDP). Algorithms such as Q-learning and DQN were used to solve the approximate optimal network task allocation solution. Drones were used to offload delay-sensitive tasks to nearby base stations or drones for calculation. The task offloading model of edge computing was designed, and the vehicle trajectory was predicted based on LTSM to increase the efficiency of task offloading. In the above solutions, no matter how the scheduling benefit is defined, they all meet user needs from the application level. This task scheduling decision can improve resource utilization efficiency and user satisfaction.

[0004] However, the existing technology still has the following shortcomings: 1. It does not effectively solve the problem of achieving seamless switching between nodes in heterogeneous network scenarios; 2. It does not comprehensively consider the needs of task offloading cost, latency, energy consumption, etc. for emergency rescue scenarios. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art or related technology, and discloses a task offloading method for emergency rescue scenarios in an integrated air-space-ground network. For a large number of tasks that need to be processed in real time in emergency rescue scenarios such as earthquake rescue and post-disaster reconstruction, reasonable and effective resource allocation is carried out, thereby improving scheduling efficiency.

[0006] The first aspect of the present invention discloses a task offloading method for emergency rescue scenarios in an integrated air-ground-space network, including: unified feature description of task offloading: using a unified description method to represent the offloading delay and node power consumption of ground servers, drones and low-orbit satellites; establishing a joint optimization problem model of delay and energy consumption: based on the offloading delay and node power consumption, a joint optimization problem model is established with the minimization of the offloading delay and power consumption as the optimization goal; designing task priority: defining the brightness value of the firefly algorithm according to the maximum tolerable delay and task size of the computing tasks generated in real time, and determining the task offloading priority based on the firefly algorithm; solving the joint optimization problem model based on a deep reinforcement learning algorithm: constructing a reward mechanism for a deep reinforcement learning model according to the task offloading priority, and using the deep reinforcement learning algorithm to solve the joint optimization problem model to obtain a variety of task offloading schemes; correcting model parameters: performing simulation verification based on the task offloading scheme to obtain evaluation data; correcting the parameters of the joint optimization problem model according to the evaluation data to obtain the optimal model for task offloading; obtaining a task offloading scheme: solving the optimal model for task offloading to obtain the final task offloading scheme.

[0007] In this technical solution, the task offloading problem is modeled as a Markov decision process required by reinforcement learning. The firefly algorithm is used to describe the urgency of the offloading task, and its quantitative index is used as a consideration of the reinforcement learning reward function, so that the agent takes actions with high reward values ​​(i.e., offloading decisions) to obtain the optimal strategy. Task scheduling can be achieved while minimizing costs and using resources as efficiently as possible.

[0008] According to the task offloading method for emergency rescue scenarios in the air-space-ground integrated network disclosed by the present invention, preferably, it also includes: constructing an air-space-ground integrated network: the air-space-ground integrated network includes: a device layer, a control layer and a management layer; the device layer includes user terminals, drones, low-orbit satellites and base stations available on the ground after the disaster; the control layer includes a network controller; the management layer includes a data processing center; the device layer and the control layer are used for data collection and transmission, and the management layer is used to generate task offloading decisions.

[0009] According to the task offloading method for emergency rescue scenarios in an air-ground integrated network disclosed by the present invention, preferably, a unified description method is used to represent the offloading delay of the ground server and the node power consumption of the ground server, specifically including:

[0010] The unloading delay model of task i to the ground server g in the disaster area is expressed as:

[0011]

[0012] in, represents the unloading delay of the ground server, Z i,g represents the local computing resources required for offloading task i, represents the computing power of the ground server when executing task i at time slot t;

[0013] The node power consumption model of the ground server when performing tasks is expressed as:

[0014] Q g,total =g i,t T i,total

[0015] Among them, Q g，total represents the node power consumption of the ground server,

[0016] g i,t It represents the power consumption of task i when it is offloading to the ground server at time slot t;

[0017] T i,total Indicates the time taken to execute the task.

[0018] According to the task offloading method for emergency rescue scenarios in an air-ground integrated network disclosed by the present invention, preferably, a unified description method is used to represent the unloading delay of the drone and the node power consumption of the drone, specifically including:

[0019] The unloading delay model of the task to the UAV in the disaster area is expressed as:

[0020] Total delay:

[0021]

[0022] Upload delay:

[0023]

[0024] Communication rate:

[0025]

[0026] Transmission costs:

[0027]

[0028] Among them, B a represents the channel transmission bandwidth of task i to the UAV, p a Indicates the user's transmit power, The channel gain between the user and the drone, σ 2 represents Gaussian white noise;

[0029] The node power consumption model of the UAV in the disaster area when performing tasks is expressed as:

[0030]

[0031] in, It represents the power of task i when the UAV is unloading the task at time slot t.

[0032] According to the task offloading method for emergency rescue scenarios in the air-ground integrated network disclosed by the present invention, preferably, a unified description method is used to represent the offloading delay of the low-orbit satellite and the node power consumption of the low-orbit satellite, specifically including:

[0033] The unloading delay model of the task to the low-orbit satellite is expressed as:

[0034] Total delay:

[0035]

[0036] Upload delay:

[0037]

[0038] Calculate the delay:

[0039]

[0040] Communication rate:

[0041]

[0042] Transmission costs:

[0043]

[0044] Among them, B s represents the channel transmission bandwidth of task i to the satellite, p s Indicates the user's transmit power, represents the channel gain between the user and the satellite, Z i,s represents the satellite computing resources required for task offloading, represents the computing power of the satellite to perform task i at time slot t;

[0045] The node power consumption model of a low-orbit satellite when performing a mission is expressed as:

[0046]

[0047] in, It represents the power of task i when the satellite is unloading the task at time slot t.

[0048] According to the task offloading method for emergency rescue scenarios in an air-ground integrated network disclosed by the present invention, preferably, the joint optimization problem model specifically includes:

[0049] The offloading decision variables of task i are defined as natural numbers a, b, c, where a+b+c=1; a=1 means the task is executed locally, b=1 means the task is executed on the drone, and c=1 means the task is executed on the satellite;

[0050] The total delay of task execution is expressed as:

[0051] T total =aT i g,total +bT i a,total +cT i s,total

[0052] The total energy consumption of task execution is expressed as:

[0053] Q total =aQ g,total +bQ a,total +cQ s,total ;

[0054] Define Z i is the task offloading cost of task i, κ and is the weight factor of delay and energy consumption, then the established joint optimization model of delay and energy consumption is as follows:

[0055]

[0056] Among them, Z i The total delay T for task offloading total and the total energy consumption of task offloading Qtotal The objective function is to minimize the total delay and total energy consumption of task offloading under the premise of not exceeding the available resources in the disaster area scenario:

[0057]

[0058] stC1:

[0059] C2:T total ≤T max

[0060] C3:

[0061]

[0062] C4:a,b,c∈{0,1}

[0063] Constraint C1 represents the value range of the weight factors of delay and energy consumption, constraint C2 represents the delay constraint when offloading tasks, constraint C3 represents the constraint on the computing power of offloading tasks to ground servers, drone servers, and low-orbit satellite servers respectively, and constraint C4 represents that the offloading decision variable is 0 or 1, a=1 means that the task is executed locally, b=1 means that the task is executed on the drone, c=1 means that the task is executed on the satellite, and a+b+c=1.

[0064] According to the task offloading method for emergency rescue scenarios in an air-ground integrated network disclosed by the present invention, preferably, the step of determining the task offloading priority based on the firefly algorithm specifically includes:

[0065] The brightness function represents the fitness of each firefly, and the brightness value is defined according to the task size and the maximum tolerable delay. i ,The larger the task volume, the more time it takes to unload, and the larger the maximum tolerable delay of the task, the less urgent the task is. The brightness value function is designed to evaluate the quality of the task priority scheme, namely:

[0066]

[0067] The position vector of each firefly represents a task priority scheme, and the distance between fireflies is expressed using the Euclidean distance formula, where n is the number of dimensions and x is q,k and x w,k Respectively represent the positions of fireflies q and w in the kth dimension, that is:

[0068]

[0069] The firefly position update rule is as follows:

[0070]

[0071] in, and ω are adjustment parameters, is the step factor, rand() generates random numbers in the range [0, 1).

[0072] According to the task offloading method for emergency rescue scenarios in an air-ground integrated network disclosed in the present invention, preferably, the step of solving the joint optimization problem model based on a deep reinforcement learning algorithm specifically includes:

[0073] Standard model based on deep reinforcement learning algorithm,

[0074] Define the state space: Make the best action based on the current amount of tasks, task priority, and server resource status. The state space is expressed as follows:

[0075]

[0076] in, They represent the remaining computing resources of drones, low-orbit satellites, and ground servers respectively. Represents the state of task i at time t, and the resource quantity Z is calculated based on the task requirements i , the distance between the task and the server d i And the task priority Mi is composed of:

[0077]

[0078] Define the action space: The action space contains the offloading decision variables a, b, and c. a = 1 means that the task is executed locally, b = 1 means that the task is offloaded to the drone for execution, and c = 1 means that the task is offloaded to the satellite for execution.

[0079] Define the state transition probability: S at time t t Take action A in state t After that, the task status and server computing resources will change, which affects the agent in state S at time t+1 t+1 Action A t+1 , use S*A*S to represent the state transition probability P of the system;

[0080] Define the reward function: Consider the delay and energy consumption requirements caused by task offloading, and introduce coefficients to integrate the influencing factors of delay and energy consumption. The ultimate goal is to minimize the cost. The higher the task priority, the higher the reward the agent gets when selecting an action. Introduce the priority attenuation factor ω, and the reward function is shown as follows:

[0081] R(S t|S t+1 ,A t )=-Z i +ωM i .

[0082] The beneficial effects of the present invention include at least: the present invention focuses on the urgency of the task, uses an optimization algorithm to quantify the priority of the task, and connects the reward mechanism of the deep reinforcement learning (DRL) model according to the priority quantification index, so as to obtain the optimal task unloading solution, which can make the task scheduling as efficient as possible while satisfying the minimum cost. Specifically, the present invention first considers the maximum tolerable delay of the task and the size of the task, and analyzes and optimizes the quantitative index of the task priority through the firefly algorithm; then the DRL model is constructed based on the task priority, and the reward function of the DRL model designed by the present invention takes into account the task unloading cost and the task priority quantitative index, and obtains the optimal unloading strategy; finally, through simulation verification and correction of model parameters, the continuous optimization of system performance is achieved. Through the present invention, the problem of lack of collaborative optimization of task unloading delay and energy consumption in the prior art can be effectively solved, the responsiveness and resource utilization efficiency of the system can be improved, and the problems such as the system falling into local optimality and difficulty in handling complex constraints caused by the prior art only using intelligent optimization algorithms to solve problems can be effectively solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A system framework diagram of a task offloading method for emergency rescue scenarios in an air-ground integrated network according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0084] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the limitations of the specific embodiments disclosed below.

[0085] An embodiment of the present invention proposes a task offloading method for emergency rescue scenarios in an air-ground integrated network. The system framework of the method in the actual application scenario is as follows: Figure 1As shown in the figure: In an integrated air-ground-space network environment, a user set will generate a large number of tasks that need to be processed. These tasks will be offloaded to the server for calculation through an air-ground communication link or a satellite-ground communication link. The task offloading method proposed in the present invention includes the joint optimization of offloading delay and energy consumption. A joint optimization model is designed and solved by a deep reinforcement learning algorithm driven by task priority, and the parameters of the joint optimization model are optimized, thereby optimizing the task offloading strategy and improving the scheduling efficiency of the task offloading process.

[0086] The task offloading method for emergency rescue scenarios in an air-ground integrated network provided in this embodiment includes:

[0087] (1) Aiming at the resource allocation problem in emergency rescue scenarios, the overall system architecture of Space-Air-Ground Integrated Networks (SAGIN) is proposed;

[0088] (2) Task priority is designed based on the maximum tolerable delay and task size of real-time tasks;

[0089] (3) Aiming at the task offloading model in emergency rescue scenarios, a task offloading scheme based on deep reinforcement learning of task priority is proposed;

[0090] (4) In a simulation environment, verify and evaluate the obtained task offloading scheme, obtain evaluation data, modify relevant parameters of the task offloading general model, and obtain the task offloading optimization model;

[0091] (5) Solve the task offloading optimization model and obtain the final task offloading solution.

[0092] According to the above embodiment, the scenario discussed in step (1) is mainly for the earthquake disaster scenario, and the resource allocation for the emergency real-time tasks generated in the process of disaster area reconstruction. The entire system is considered to be a network area under a certain coverage range of the satellite. SAGIN consists of three layers: equipment layer, control layer and management layer. The equipment layer includes all communication nodes within the coverage range of the system, such as base stations, unmanned aerial vehicles (UAVs), and low earth orbit satellites (LEOs) that can be used on the ground after the disaster. The network controller of the control layer controls the respective underlying physical resources. The management layer is mainly used to coordinate different network allocation actions, and is equipped with a data processing center, which generates unloading decisions. The equipment layer and the control layer are used as the data acquisition layer and the transmission layer, and the optimal network task allocation plan is given based on the management layer function.

[0093] Preferably, step (1) specifically includes:

[0094] (1a) The set of communication nodes is K, where communication node k j ∈K;

[0095] (1b) Assume that the set of users in the system is U, and user u j ∈U;

[0096] (1c) The task generated by user u in the system is i, and the task set is I, where i j ∈I, the task size is denoted as h i , the maximum tolerable delay of task i is T i,max , where j=1,2,3,4...

[0097] (1d) Taking into account the saving of transmission resources and reducing communication congestion, a number of available servers will still be retained on the ground after a disaster occurs, and tasks can be executed locally, that is, the tasks are offloaded to the available servers in the disaster area for execution.

[0098] The local processing delay of task i is as follows, where Z i,g Indicates the local computing resources required for task offloading, represents the computing power of the ground server when executing task i at time slot t,

[0099]

[0100] The energy consumed by executing the task locally (on the ground server) is as follows: i,t It represents the power of task i when it is offloading locally at time slot t:

[0101] Q g,total =g i,t T i g,total

[0102] (1e) UAVs play an important role in disaster relief work. They can be deployed in places where rescue is inconvenient and where communication resources are limited. Therefore, a large number of real-time tasks generated in disaster areas need to be offloaded to UAVs for processing, or UAVs can be used as relay nodes to transmit to low-orbit satellites.

[0103] The delay caused by unloading task i to the drone mainly includes task upload and task processing delay. The communication rate between task i and the drone is expressed as follows, where: B a represents the channel transmission bandwidth of task i to the UAV, p a Indicates the user's transmit power, The channel gain between the user and the drone, σ 2 represents Gaussian white noise. :

[0104]

[0105] The transmission cost between the user and the UAV for task i in time slot t is defined as follows:

[0106]

[0107] in, represents the distance between the user and the drone at time slot t;

[0108] Then the delay of uploading task i to the drone in time slot t is as follows:

[0109]

[0110] Among them, G i Indicates the size of the task i;

[0111] The execution delay of task i on the drone is as follows:

[0112]

[0113] Among them, Z i,a represents the UAV computing resources required for task offloading, represents the computing power of the UAV to perform task i at time slot t;

[0114] In summary, the delay caused by offloading task i to the drone is as follows:

[0115]

[0116] The energy consumed by the drone during the task execution is as follows:

[0117]

[0118] in It represents the power of task i when the UAV is unloading the task at time slot t;

[0119] (1f) The delay caused by task i being unloaded to the satellite mainly includes task uploading, task processing, and task processing result feedback. In order to simplify the model, the delay caused by task processing result feedback is negligible. The communication rate between task i and the satellite can be expressed as follows: s represents the channel transmission bandwidth of task i to the satellite, p s Indicates the user's transmit power, Represents the channel gain between the user and the satellite:

[0120]

[0121] Definition c i,s represents the transmission cost between the user and the satellite for task i in time slot t, as follows:

[0122]

[0123] in, It is represented as the distance between the user and the satellite at time slot t;

[0124] Then the time delay of task i being uploaded to the satellite in time slot t is as follows:

[0125]

[0126] Among them, G i Indicates the size of the task i;

[0127] The execution delay of task i on the satellite is as follows:

[0128]

[0129] Among them, Z i,s represents the satellite computing resources required for task offloading, represents the computing power of the satellite to perform task i at time slot t;

[0130] In summary, the delay caused by unloading task i to the satellite for execution is as follows:

[0131]

[0132] The energy consumed by the satellite during mission execution is as follows:

[0133]

[0134] in represents the power of task i when the satellite is unloading the task at time slot t;

[0135] (1g) Define the offloading decision variables of task i as natural numbers a, b, c, where a+b+c=1. a=1 means the task is executed locally, b=1 means the task is executed on the drone, and c=1 means the task is executed on the satellite. The total delay of task execution is expressed as follows:

[0136] T total =aT i g,total +bT i a,total +cT i s,total

[0137] The total energy consumption of task execution is shown as follows:

[0138] Q total =aQ g,total +bQ a,total+cQ s,total

[0139] The present invention takes the minimization of the unloading delay and energy consumption of the joint task as the optimization goal and defines Z i is the task offloading cost of task i, κ and is the weight factor of delay and energy consumption, then the established joint optimization model of delay and energy consumption is as follows:

[0140]

[0141] The objective function is constructed as shown below:

[0142]

[0143] stC1:

[0144] C2:T total ≤T max

[0145] C3:

[0146]

[0147] C4:a,b,c∈{0,1}

[0148] Where: Z i The total delay T for task offloading total and the total energy consumption of task offloading Q total The objective function is to minimize the total delay and total energy consumption of task offloading in the disaster area under the premise of not exceeding the available resources. C1 represents the weight factor of delay and energy consumption; constraint C2 represents the delay constraint when offloading tasks; each formula of constraint C3 represents the constraints on the computing power of offloading tasks to ground, air, and space servers respectively. They represent the maximum computing capabilities of the ground server, UAV server, and low-orbit satellite server respectively; constraint C4 indicates that the offloading decision variable is an integer in {0, 1}, and if a, b, and c are 1 respectively, it means that the task is executed locally, offloaded to the UAV, or offloaded to the satellite.

[0149] According to the above embodiment, preferably, step (2) specifically includes: after the earthquake, different tasks of different users have different requirements for network resources, and the description of the user's urgency involves many key factors. In order to solve the problem of network task offloading within a certain coverage area of ​​the satellite, the following key steps are mainly required: data collection for real-time tasks in the area, and then overall data analysis, including user configuration, resource requirements, etc., and finally resource allocation plans are adopted according to the user's urgency. A set of resources available to users is established through the network allocation function, with reference to the behavior of fireflies, which attract each other based on their own brightness and attractiveness. In the task scheduling system, factors such as task size and maximum tolerable delay of the task are introduced to derive the task priority, and the firefly algorithm is used to calculate the task size h of a known task. i and the maximum tolerable delay T i,max , get the priority M of the task i , its time complexity is O(m 2 ), and because the location information of fireflies needs to be stored, the space complexity of the algorithm is O(m), where m is the number of fireflies. The algorithm initially randomly defines a brightness for each firefly, i.e., fitness. And generates a set of firefly positions, each of which represents a solution vector, i.e., the priority quantization scheme of the input task. Then, "designing the task priority according to the maximum tolerable delay and task size of the real-time task" specifically includes the following steps:

[0150] (2a) The brightness function represents the fitness of each firefly. The brightness value can be defined according to the task size and the maximum tolerable delay. i , the larger the task volume, the more time it takes to unload, which may cause the waiting time of tasks with small task volume to be too long. The larger the maximum tolerance delay of the task, the less urgent the task is. Therefore, the brightness value function is designed to evaluate the quality of the task priority scheme, that is:

[0151]

[0152] (2b) The position vector of each firefly represents a task priority scheme. The distance between fireflies can be expressed using the Euclidean distance formula, where n is the number of dimensions and x is q,k and x w,k Respectively represent the positions of fireflies q and w in the kth dimension, that is:

[0153]

[0154] (2c) The position update rule of the firefly at time step t+1 is as follows, where It represents the position of the firefly in the qth dimension after the update at time t+1. represents the position of the firefly in the qth dimension at time t represents the position of the firefly in the wth dimension at time t. and ω are adjustment parameters, is the step factor, rand() generates random numbers in the range [0, 1):

[0155]

[0156] The firefly algorithm can help determine the priority of tasks so that task scheduling can utilize resources as efficiently as possible while meeting the maximum tolerable delay.

[0157] According to the above embodiment, preferably, step (3) specifically includes: in the earthquake rescue and post-disaster reconstruction scenario, the tasks generated by the user can select multiple offloading targets. In the air-ground integrated architecture, the problem of offloading tasks to the appropriate optimal server needs to be solved urgently. In order to solve the problem that offloading decisions are difficult to optimize, the present invention proposes a deep reinforcement learning algorithm based on task priority (TPODRL, Deep Reinfoece Learning Base on Task Priority) on the basis of quantifying the urgency of the task, and designs a task offloading scheme:

[0158] The generation of disaster area tasks and their offloading to the target server for processing is a random process. In order to reduce the problem of dimensional explosion of optimization problems caused by various complexities, the standard model MDP of reinforcement learning is used in the SAGIN architecture scenario. MDP makes the acquisition of rewards independent of past states and actions through a high degree of abstraction of the real world, which can greatly reduce the complexity of the problem. Through the above modeling, the present invention abstracts the problem into an MDP quadruple, which allows the intelligent agent to explore the optimal offloading strategy in an established environment. MDP can be represented by (S, A, P, R). Among them, S represents the state space, A represents the action space, P represents the state transition probability, and R represents the reward.

[0159] State space: The CPU needs to make the best action based on the current task size, task priority, server resource status and other conditions. At time slot t, the state can be expressed as follows:

[0160]

[0161] in, Represents the state of task i at time t, and the resource quantity Z is calculated based on the task requirements i , the distance between the task and the server d i and the task priority Mi, Respectively represent the remaining computing resources of the air, sky, and ground servers:

[0162]

[0163] Action space: Action space refers to the task offloading plan in each time slot combined with the current state of the generated task. The action space contains the offloading decision variables a, b, and c. a = 1 means that the task is executed locally, b = 1 means that the task is offloaded to the drone for execution, and c = 1 means that the task is offloaded to the satellite for execution.

[0164] State transition probability: S at time t t Take action A in state t After that, the task status and server computing resources will change, which affects the agent in state S at time t+1 t+1 Action A t+1 , use S*A*S to represent the state transition probability P of the system.

[0165] Reward function: The entire system takes into account the delay and energy consumption requirements caused by task offloading, and introduces coefficients to integrate the influencing factors of delay and energy consumption. The ultimate goal of the system is to minimize the cost, and the higher the task priority, the higher the reward the agent gets when choosing an action. The priority attenuation factor ω is introduced, and the reward function is shown as follows:

[0166] R(S t |S t+1 ,A t )=-Z i +ωM i

[0167] In summary, the present invention provides a task offloading strategy in emergency rescue scenarios, which mainly includes a task transmission model and a task execution model. Taking into account the data volume and the maximum tolerable delay of the task, the firefly algorithm is used to divide the priority of the task, and a task offloading plan is formulated on this basis, which can effectively alleviate the problem of excessive load on the ground server after the disaster area, and can greatly save rescue time and reduce system energy consumption.

[0168] All or part of the steps in the various methods of the above embodiments can be completed by controlling the relevant hardware through a program, and the program can be stored in a readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other readable medium that can be used to carry or store data.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A task offloading method for emergency rescue scenarios in an air-ground integrated network, characterized in that: include: Unified feature description of task offloading: Use a unified description method to represent the offloading delay and node power consumption of ground servers, drones, and low-orbit satellites; Establishing a joint optimization problem model of delay and energy consumption: Based on the unloading delay and the node power consumption, a joint optimization problem model is established with minimization of the unloading delay and power consumption as the optimization goal; Design task priority: define the brightness value of the firefly algorithm according to the maximum tolerable delay and task size of the computing task generated in real time, and determine the task offloading priority based on the firefly algorithm; Solving the joint optimization problem model based on a deep reinforcement learning algorithm: constructing a reward mechanism of a deep reinforcement learning model according to the task offloading priority, solving the joint optimization problem model using a deep reinforcement learning algorithm, and obtaining a variety of task offloading solutions; Correcting model parameters: performing simulation verification based on the task offloading scheme to obtain evaluation data; Modifying the parameters of the joint optimization problem model according to the evaluation data to obtain an optimal model for task offloading; Obtaining a task offloading solution: solving the task offloading optimal model to obtain a final task offloading solution.

2. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 1 is characterized in that: Also includes: Build an integrated air-space-ground network: The integrated air-space-ground network includes: an equipment layer, a control layer and a management layer; the equipment layer includes user terminals, drones, low-orbit satellites and base stations available on the ground after a disaster; the control layer includes a network controller; the management layer includes a data processing center; the equipment layer and the control layer are used for data collection and transmission, and the management layer is used to generate task offloading decisions.

3. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 1 is characterized in that: A unified description method is used to represent the unloading delay of the ground server and the node power consumption of the ground server, specifically including: The unloading delay model of task i to the ground server g in the disaster area is expressed as: Among them, T i g，total represents the unloading delay of the ground server, Z i,g represents the local computing resources required for offloading task i, represents the computing power of the ground server when executing task i at time slot t; The node power consumption model of the ground server when performing tasks is expressed as: Q g,total =g i,t T i,total Among them, Q g，total represents the node power consumption of the ground server, g i,t It represents the power consumption of task i when it is offloading to the ground server at time slot t; T i,total Indicates the time taken to execute the task.

4. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 3 is characterized in that: A unified description method is used to represent the unloading delay of the drone and the node power consumption of the drone, specifically including: The unloading delay model of the task to the UAV in the disaster area is expressed as: Total delay: Upload delay: Calculate the delay: Communication rate: Transmission costs: Among them, B a represents the channel transmission bandwidth of task i to the UAV, p a Indicates the user's transmit power, The channel gain between the user and the drone, σ 2 represents Gaussian white noise, Z i,a represents the UAV computing resources required for task offloading, represents the computing power of the UAV to perform task i at time slot t; The node power consumption model of the UAV in the disaster area when performing tasks is expressed as: in, It represents the power of task i when the UAV is unloading the task at time slot t.

5. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 4 is characterized in that: A unified description method is used to represent the unloading delay of the low-orbit satellite and the node power consumption of the low-orbit satellite, specifically including: The unloading delay model of the task to the low-orbit satellite is expressed as: Total delay: Upload delay: Calculate the delay: Communication rate: Transmission costs: Among them, B s represents the channel transmission bandwidth of task i to the satellite, p s Indicates the user's transmit power, represents the channel gain between the user and the satellite, Z i,s represents the satellite computing resources required for task offloading, represents the computing power of the satellite to perform task i at time slot t; The node power consumption model of a low-orbit satellite when performing a mission is expressed as: in, It represents the power of task i when the satellite is unloading the task at time slot t.

6. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 5 is characterized in that: The joint optimization problem model specifically includes: The offloading decision variables of task i are defined as natural numbers a, b, c, where a+b+c=1; a=1 means the task is executed locally, b=1 means the task is executed on the drone, and c=1 means the task is executed on the satellite; The total delay of task execution is expressed as: T total =aT i g,total +bT i a,total +cT i s,total The total energy consumption of task execution is expressed as: Q total =aQ g,total +bQ a,total +cQ s,total ; Define Z i is the task offloading cost of task i, κ and is the weight factor of delay and energy consumption, then the established joint optimization model of delay and energy consumption is as follows: Among them, Z i The total delay T for task offloading total and the total energy consumption of task offloading Q total The objective function is to minimize the total delay and total energy consumption of task offloading under the premise of not exceeding the available resources in the disaster area scenario: s.t.C1: C2:T total ≤T max C3: C4:a,b,c∈{0,1} Constraint C1 represents the value range of the weight factor of delay and energy consumption, constraint C2 represents the delay constraint when offloading tasks, and constraint C3 represents the constraint on the computing power of offloading tasks to ground servers, drone servers, and low-orbit satellite servers. They represent the maximum computing capabilities of the ground server, UAV server, and low-orbit satellite server respectively. The constraint C4 indicates that the offloading decision variable is 0 or 1, a=1 means that the task is executed locally, b=1 means that the task is executed on the UAV, c=1 means that the task is executed on the satellite, and a+b+c=1.

7. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 1 is characterized in that: The step of determining the task offloading priority based on the firefly algorithm specifically includes: The brightness function represents the fitness of each firefly, and the brightness value is defined according to the task size and the maximum tolerable delay. i ,The larger the task volume, the more time it takes to unload, and the larger the maximum tolerable delay of the task, the less urgent the task is. The brightness value function is designed to evaluate the quality of the task priority scheme, namely: The position vector of each firefly represents a task priority scheme, and the distance between fireflies is expressed using the Euclidean distance formula, where n is the number of dimensions and x is q,k and x w,k Respectively represent the positions of fireflies q and w in the kth dimension, that is: The firefly position update rule is as follows: in, and ω are adjustment parameters, θ is the step factor, and rand() generates random numbers in the range of [0, 1).

8. The task offloading method for emergency rescue scenarios in an air-ground integrated network according to claim 1 is characterized in that: The step of solving the joint optimization problem model based on the deep reinforcement learning algorithm specifically includes: Standard model based on deep reinforcement learning algorithm, Define the state space: Make the best action based on the current amount of tasks, task priority, and server resource status. The state space is expressed as follows: in, They represent the remaining computing resources of drones, low-orbit satellites, and ground servers respectively. Represents the state of task i at time t, and the resource quantity Z is calculated based on the task requirements i , the distance between the task and the server d i And the task priority Mi is composed of: Define the action space: The action space contains the offloading decision variables a, b, and c. a = 1 means that the task is executed locally, b = 1 means that the task is offloaded to the drone for execution, and c = 1 means that the task is offloaded to the satellite for execution. Define the state transition probability: S at time t t Take action A in state t After that, the task status and server computing resources will change, which affects the agent in state S at time t+1 t+1 Action A t+1 , use S*A*S to represent the state transition probability P of the system; Define the reward function: Consider the delay and energy consumption requirements caused by task offloading, and introduce coefficients to integrate the influencing factors of delay and energy consumption. The ultimate goal is to minimize the cost. The higher the task priority, the higher the reward the agent gets when selecting an action. Introduce the priority attenuation factor ω, and the reward function is shown as follows: R(S t |S t+1 ,A t )=-Z i +ωM i 。

Images