WO2023245835A1 - 一种基于栅格时延预测的卫星传输优化方法 - Google Patents

一种基于栅格时延预测的卫星传输优化方法 Download PDF

Info

Publication number
WO2023245835A1
WO2023245835A1 PCT/CN2022/112499 CN2022112499W WO2023245835A1 WO 2023245835 A1 WO2023245835 A1 WO 2023245835A1 CN 2022112499 W CN2022112499 W CN 2022112499W WO 2023245835 A1 WO2023245835 A1 WO 2023245835A1
Authority
WO
WIPO (PCT)
Prior art keywords
satellite
user
time slot
delay
task
Prior art date
Application number
PCT/CN2022/112499
Other languages
English (en)
French (fr)
Inventor
丁飞
马海蓉
庄衡衡
张美楠
马文
Original Assignee
南京邮电大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京邮电大学 filed Critical 南京邮电大学
Publication of WO2023245835A1 publication Critical patent/WO2023245835A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/1851—Systems using a satellite or space-based relay
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/1851—Systems using a satellite or space-based relay
    • H04B7/18513—Transmission in a satellite or space-based system
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/18521—Systems of inter linked satellites, i.e. inter satellite service
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/18578—Satellite systems for providing broadband data service to individual earth stations
    • H04B7/18584—Arrangements for data networking, i.e. for data packet routing, for congestion control
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/18578—Satellite systems for providing broadband data service to individual earth stations
    • H04B7/18586—Arrangements for data transporting, e.g. for an end to end data transport or check
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00—Routing or path finding of packets in data switching networks
    • H04L45/302—Route determination based on requested QoS
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/1021—Server selection for load balancing based on client or server locations
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Definitions

  • the invention belongs to the technical field of satellite transmission task offloading and resource allocation, and in particular relates to a satellite transmission optimization method based on grid delay prediction.
  • the purpose of the present invention is to provide a satellite transmission optimization method based on grid delay prediction, aiming at the time-varying characteristics caused by satellite motion, the sudden arrival of satellite network services, and the randomness of system services.
  • a satellite IoT business modeling analysis method based on geographical raster and random network calculation, deriving the grid delay boundary as a Qos service quality guarantee, and jointly considering task offloading decisions and scheduling decisions based on the grid statistical delay guarantee.
  • communication and computing resource allocation method the optimization problem is formulated as a mixed integer nonlinear programming problem. In order to solve this problem, it is decoupled into two sub-problems: fair resource allocation based on grid delay guarantee and task offloading and scheduling decision-making. , using the dual ascent method and the deep Q network (DQN) to solve the two sub-problems respectively to improve the resource utilization of the satellite network.
  • DQN deep Q network
  • the present invention adopts the following technical solutions:
  • the present invention provides a satellite transmission optimization method, including:
  • the grid delay boundary is derived, and the grid delay limit value is set based on the grid delay boundary as a grid delay guarantee ;
  • the optimization problem is formulated as a mixed integer nonlinear programming problem P1;
  • the dual ascent method is used to solve the fair communication and computing resource allocation sub-problems, and fair communication resource allocation and computing resource allocation are obtained;
  • Deep Q Network is used to solve the task offloading and scheduling decision sub-problems, and the offloading decision and scheduling decision of the user's current task are obtained.
  • w is the total number of user tasks completed and returned results within l time slots; when the task k of user m within the range of grid i is within the limited delay set based on SNC derivation, Completed within and return the result, i.e.
  • Constraint (18a) ensures that the communication resources allocated by the satellite in the lth time slot to user m It should not be larger than the communication resources that can be allocated by satellite n in the lth time slot.
  • Constraint (18b) ensures that the computing resources allocated by the satellite in the lth time slot to user m It should not be larger than the computing resources that can be allocated by satellite n in the lth time slot.
  • Constraint (18c) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings
  • the transmission delay generated after the task of user m in the range of grid i is scheduled from the transmission queue of satellite n and communication resources are allocated;
  • C l ⁇ C 1,l ,C 2,l ,...,C N,l ⁇ , It is the communication resource allocated by the satellite to each user in the lth time slot;
  • X l ⁇ X 1,l ,X 2,l ,...,X N,l ⁇ ,
  • the computing resources allocated to each user by the satellite in the lth time slot and are respectively the communication resources and computing resources that can be allocated by satellite n in the lth time slot;
  • Constraint (22a) ensures that the communication resources allocated by the lth time slot satellite n to user m It should not be larger than the communication resources that can be allocated by satellite n in the lth time slot.
  • Constraint (22b) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings
  • Constraint (23a) ensures that the computing resources allocated by the satellite in the lth time slot to user m It should not be larger than the computing resources that can be allocated by satellite n in the lth time slot.
  • Constraint (23b) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings
  • the dual ascent method is used to solve the fair communication and computing resource allocation sub-problems, including:
  • the dual ascent method is used to solve the communication resource allocation sub-problem:
  • ⁇ m ⁇ 0, ⁇ ⁇ 0, ⁇ ⁇ 0 are Lagrange multipliers
  • the maximum value of D is the minimum value obtained by formula (22);
  • ⁇ m ⁇ 0, ⁇ ⁇ 0, ⁇ ⁇ 0 are Lagrange multipliers
  • the maximum value of D is the minimum value obtained by formula (23);
  • deep Q networks are used to solve task offloading and scheduling decision sub-problems, including:
  • IoT terminal information P U (l) , X U,l , T l , satellite information: P s (l), ⁇ l , C l , pro,l inputs the pre-trained deep Q network;
  • C l ⁇ C 1,l , C 2,l ,..., C N, l ⁇ is the communication resource occupied by the l-th time slot satellite;
  • X l ⁇ X 1,l , X 2,l ,... , ; and are the total number of tasks waiting to be scheduled in the transmission queue and processing queue of the satellite in the lth time slot respectively; It is the total number of tasks waiting for scheduling in the local user task arrival queue in the lth time slot;
  • E off ⁇ 1,2...,Z ⁇ is the offloading decision of the task
  • E exe ⁇ 0,1 Indicates whether to schedule tasks from the satellite's transmission or processing queue; the offloading decision of the user's current task can be obtained through action a l and scheduling decisions
  • the grid delay boundary is expressed as:
  • D(n) is the delay function
  • P(D(n)>x) represents the probability that the delay exceeds a certain value x
  • ⁇ (n) is the random service curve
  • the present invention provides a satellite transmission optimization device, including a processor and a storage medium;
  • the storage medium is used to store instructions
  • the processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.
  • the present invention provides a storage medium on which a computer program is stored.
  • the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.
  • the present invention sets the task limit delay based on the grid delay boundary derived by random network calculation.
  • the grid resource constraint of the satellite it can guarantee the grid service QoS. , realize fast, accurate, low-overhead evaluation and inference of routing key performance indicators (KPI) in dynamic network environments on the basis of user fairness, and improve the resource utilization of satellite networks.
  • KPI routing key performance indicators
  • Figure 1 is a flow chart of a satellite transmission optimization method based on grid delay prediction in an embodiment of the present invention.
  • Figure 2 is a flow chart of grid business modeling in the embodiment of the present invention.
  • Figure 3 is an architecture diagram of a converged satellite-ground collaborative network in an embodiment of the present invention.
  • Figure 4 is a flow chart of the joint algorithm in the embodiment of the present invention.
  • a satellite transmission optimization method including:
  • the grid delay boundary is derived, and the grid delay boundary is used as a service quality guarantee; the grid is set based on the grid delay boundary
  • the delay limit value serves as the grid delay guarantee; (the limited delay set for task k of user m within the range of grid i based on SNC random network calculation derivation is expressed as
  • the optimization problem is formulated as a mixed integer nonlinear programming problem P1;
  • the dual ascent method is used to solve the fair communication and computing resource allocation sub-problems, and fair communication resource allocation and computing resource allocation are obtained;
  • Deep Q Network is used to solve the task offloading and scheduling decision sub-problems, and the offloading decision and scheduling decision of the user's current task are obtained.
  • a satellite transmission optimization method based on grid delay prediction includes:
  • Step 1 Network business modeling and analysis based on geographical raster and random network calculation:
  • the services generated by the terminal may be aggregated through different satellite nodes each time.
  • the motion trajectory of the satellite sub-satellite point can be calculated from the satellite ephemeris.
  • This solution uses stochastic network calculus theory combined with the geographical grid method to analyze the queuing performance of the satellite network.
  • the service arrival and channel service model of the satellite network sufficient Consider the randomness and burstiness of satellite network service arrivals, as well as the randomness of system services introduced by multiple access protocols.
  • the research ideas are as follows:
  • the earth's surface is unfolded into a two-dimensional plane, and the earth's surface is divided into many grids according to certain longitude and latitude intervals. Since the divided grids may contain a variety of geographical environments, the equipment deployment density of a certain grid Determined by a variety of geographical environments, the device deployment density of grid i is defined as:
  • S i represents the area of grid i
  • n represents the number of geographical environment types contained in the grid
  • ⁇ j represents the device deployment density corresponding to geographical environment type j
  • S i,j represents the geographical environment type j in grid i.
  • the area occupied. Si is calculated as:
  • ⁇ is the longitude range of the grid
  • a e is the radius of the earth's equatorial circle
  • e 1 is the first eccentricity
  • the projection of the satellite beam on the ground will directly determine its coverage, thereby determining the number of grids within its coverage area. Since IoT services converge at the satellite level, in order to simplify the calculation of coverage area, all point beams of the satellite can be regarded as a synthetic beam, and its projection on the ground can be approximately regarded as the area surrounded by a circle with the sub-satellite point as the center. , as shown in the figure, its central angle is:
  • Re is the radius of the earth
  • H is the orbital height
  • ⁇ is the minimum communication angle
  • Stochastic network calculus theory is a new queuing theory analysis method. This theory does not limit the business arrival and service service in the network to a specific distribution, but uses the envelope method to model the business arrival and channel service in the network, and Introducing violation probability into the model can accurately describe the randomness and burstiness of services in satellite networks and the randomness of channel services.
  • the Markov-modulated ON-OFF (MMOO) process is used to The model is characterized by service arrival. This model has great adjustability and can be applied to different levels of burst services.
  • State 1 means that the sending node is in the ON state and sends data packets at a fixed rate h; state 0 means that the sending node is in the OFF state and no data packets are generated at this time.
  • the probability of transitioning from state 0 to state 1 is ⁇ , and the probability of transitioning from 1 to 0 is ⁇ .
  • the satellite network service arrival process A(n) has a vbc random arrival curve, and the service arrival A(n) represented by the Markov modulated switching process, have
  • the satellite network service arrival process A(n) has a vbc random arrival curve, that is: A(n) ⁇ sac ⁇ f,a>, where have
  • ⁇ (n) is a linear function with respect to n
  • ⁇ ⁇ ( ⁇ ) is its slope
  • f(x) is the boundary function of ⁇ (n).
  • the random service process S(n) of the satellite network has a random service curve S(n) ⁇ s-ssc ⁇ g, ⁇ >, where:
  • ⁇ (n) is a linear function with respect to n
  • ⁇ ⁇ ( ⁇ ) is its slope
  • g(x) is the boundary function of ⁇ (n).
  • the satellite network delay boundary expression is:
  • D(n) is the delay function
  • P(D(n)>x) represents the probability that the delay exceeds a certain value x
  • ⁇ (n) is the random service curve
  • Step 2 Task scheduling and resource allocation strategy based on grid statistical delay guarantee in the integrated MEC scenario
  • the dual ascent method is used to solve the optimal communication and computing resource allocation. Then, the results of the resource allocation sub-problem are input into the offloading decision problem, the offloading decision problem with dynamic constraints is formulated as a Markov decision process (MDP), and a deep Q network (DQN) is used to increase the long-term returns of the offloading decision. , improve the task completion rate.
  • MDP Markov decision process
  • DQN deep Q network
  • the task can be offloaded to other surrounding satellites for processing through the inter-satellite wireless link, and the additional propagation delay of about 10ms is exchanged for lower processing delay to achieve inter-satellite collaboration. processing optimization effect.
  • the collaborative processing of ground user tasks by satellite networks is realized through inter-satellite wireless links.
  • task offloading decisions, scheduling decisions, communication and computing are required.
  • Resource allocation is jointly optimized. Therefore, the optimization problem of the present invention can be expressed as,
  • w is the total number of user tasks completed and returned results within l time slots; when the task k of user m within the range of grid i is within the limited delay set based on SNC derivation, Completed within and return the result, i.e.
  • Constraint (18a) ensures that the communication resources allocated by the satellite in the lth time slot to user m It should not be larger than the communication resources that can be allocated by satellite n in the lth time slot.
  • Constraint (18b) ensures that the computing resources allocated by the satellite in the lth time slot to user m It should not be larger than the computing resources that can be allocated by satellite n in the lth time slot.
  • Constraint (18c) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings
  • the grid service limit delay is derived based on geographical grid division and random network modeling. Then in each scheduling time slot, it is judged whether satellite cutting occurs due to the movement of the satellite. Then, based on DQN's task offloading and scheduling algorithm offloads and schedules continuously arriving tasks, so that for tasks scheduled at the same time in the transmission/processing queue of the same satellite, a fair resource allocation algorithm based on the dual ascent method allocates communication and Computing resources, that is, under the premise of known task offloading and scheduling decisions, the allocation problem of communication and computing resources is modeled as a maximum-minimum fairness problem to solve to improve the system task completion rate. Repeat the above steps while the system is running.
  • ⁇ wait represents the waiting delay caused by local computing resources being occupied by tasks being processed
  • ⁇ process T m /X m
  • X m is the computing resource of user m
  • T m is the size of task k of user m.
  • ⁇ off is the delay of task offloading from the user to the access star n a , including three parts: transmission waiting delay ⁇ wait , transmission delay ⁇ trans and propagation delay ⁇ prop .
  • c is the speed of light and is the propagation distance from user m to access star n a ;
  • ⁇ process consists of two parts: waiting processing delay and processing delay.
  • the computing resources can be allocated to the processing queue only after the task transmission is completed.
  • the allocation is independent and sequential, and may not necessarily be completed in the same time slot; ⁇ return is the result return delay.
  • the shortest route distance from the access star n a to the switching star n h can be obtained by Dijkstra's algorithm, is the propagation distance from switching star n h to user m.
  • ⁇ off is the time delay for the task to be offloaded from the user m to the access star n a , and the calculation method is the same as (20); ⁇ ISL is the time delay for the task to be offloaded from the access star n a to the processing star n p through the inter-satellite link.
  • Delay considering the sufficient communication resources between satellites, the transmission delay of the inter-satellite link can be ignored, so It is the inter-satellite routing distance (single hop or multi-hop) from the access star n a to the processing star n p , which is obtained by Dijkstra's algorithm; ⁇ process includes the waiting delay of the task in the processing queue of the processing star ⁇ w_pro and the processing delay ⁇ process has two parts; ⁇ return is the propagation delay of the processing result from the processing star n p to the user m. First, the processing result will be returned from the processing star n p to the access star n a , and then the access star will return it to the user. If The processing results can be returned to the user before the access star ends coverage. otherwise, The inter-satellite routing distance from processing star n p to switching star n h .
  • the projection of satellite beams on the ground will directly determine its coverage area, thereby determining the number of grids in its coverage area. Since the system needs to provide seamless coverage around the world, satellite coverage areas will partially overlap during system design. In the overlapping area, the terminal will give priority to satellite nodes for transmission based on the principle of closest distance.
  • the satellite coverage area is divided into several priority grids according to the distance from the sub-satellite point. Within the satellite overlapping coverage area, all IoT device nodes in the same geographical grid select the satellite with the highest priority for convergence in real time. When the communication angle is less than the minimum coverage elevation angle ⁇ min , it means that the satellite is about to end covering the terminal, and a satellite with a higher priority is selected for switching based on the priority criterion again. If the task processing results cannot be returned before the access satellite ends coverage, the results will be returned through the switched satellite.
  • the priority of a certain satellite relative to the geographical grid is calculated according to the following steps:
  • Step 1 Get the latitude and longitude coordinates of the satellite sub-satellite point and determine its grid index in the geographical grid.
  • Step 2 Set the priority of the grid where the satellite sub-satellite point is located to 1, and set the corresponding priority for the surrounding grid. The closer to this grid, the smaller the distance from the satellite, the higher the priority. high.
  • Step 3 The grid takes out the satellite number with the highest priority relative to itself for access.
  • the communication and computing resource allocation problem is modeled as a max-min fairness problem to solve, in order to minimize the maximum delay of the task and improve the overall task completion rate of the system.
  • the detailed solution and derivation process is shown below.
  • the transmission delay generated after the task of user m in the range of grid i is scheduled from the transmission queue of satellite n and communication resources are allocated;
  • C l ⁇ C 1,l ,C 2,l ,...,C N,l ⁇ , It is the communication resource allocated by the satellite to each user in the lth time slot;
  • X l ⁇ X 1,l ,X 2,l ,...,X N,l ⁇ ,
  • Constraint (22a) ensures that the communication resources allocated by the satellite in the lth time slot to user m It should not be larger than the communication resources that can be allocated by satellite n in the lth time slot.
  • Constraint (22b) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings In the same way, constraint equation (23a) ensures that the computing resources allocated by the satellite in the lth time slot to user m It should not be larger than the computing resources that can be allocated by satellite n in the lth time slot.
  • Constraint (23b) ensures the link delay of user m Should not be greater than the limited delay of grid i based on SNC settings
  • the dual ascent method is used to solve the communication resource allocation sub-problem:
  • ⁇ m ⁇ 0, ⁇ ⁇ 0, ⁇ ⁇ 0 are Lagrange multipliers.
  • the dual ascent method is used to solve the computing resource allocation sub-problem, including:
  • ⁇ m ⁇ 0, ⁇ ⁇ 0, ⁇ ⁇ 0 are Lagrange multipliers
  • the maximum value of D is the minimum value obtained by formula (23);
  • the fairest allocation of communication and computing resources for each time slot can be obtained from the above.
  • the joint decision-making of offloading and scheduling is still a non-convex dynamic programming problem. Therefore, the DQN algorithm is used to solve the coupling offloading, scheduling, and resource allocation problems of multiple batches of tasks that continuously arrive over a long period of time.
  • the specific MDP representation of this problem is as follows:
  • C l ⁇ C 1,l , C 2,l ,..., C N, l ⁇ is the communication resource occupied by the l-th time slot satellite;
  • X l ⁇ X 1,l , X 2,l ,... , ; and are the total number of tasks waiting to be scheduled in the transmission queue and processing queue of the satellite in the lth time slot respectively; It is the total number of tasks waiting for scheduling in the local user task arrival queue in the lth time slot.
  • E exe ⁇ ⁇ 0,1 ⁇ indicates whether to schedule from the satellite's transmission or processing queue Task.
  • the offloading decision of the user's current task can be obtained through a l and scheduling decisions
  • the reward of selecting action a l in the l-th time slot state h l is defined as:
  • R P is a positive constant, so that d is the number of tasks completed within the limited delay in the lth time slot; R d is the additional reward for tasks completed within the limited delay.
  • Deep Q network is used to solve task offloading and scheduling decision-making sub-problems, including:
  • IoT terminal information P U (l) , X U,l , T l , satellite information: P s (l), ⁇ l , C l , pro,l inputs the pre-trained deep Q network;
  • ⁇ , ⁇ , ⁇ , and ⁇ are the network parameters involved in the training optimization process: ⁇ is the discount factor of the action value function; ⁇ is the greedy strategy selection action probability (greedy The strategy is to select a random action with a small probability and the optimal action with a high probability), that is, randomly select an action with a probability of ⁇ , and select an optimal action with a probability of 1- ⁇ ; ⁇ is the experience replay pool; ⁇ is used to calculate the training loss function , whose training goal is to minimize the loss function.
  • Task scheduling is performed once per time slot, and each time slot is a step in the training process.
  • the action is selected according to the ⁇ greedy strategy, that is, the action is randomly selected with the probability of ⁇ , and the optimal action is selected with the probability of 1- ⁇ .
  • the satellite communication resource allocation C l and the computing resource allocation X l calculated according to Algorithm 1-1 and Algorithm 1-2.
  • this embodiment provides a satellite transmission optimization device, including a processor and a storage medium;
  • the storage medium is used to store instructions
  • the processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.
  • this embodiment provides a storage medium on which a computer program is stored.
  • the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.
  • embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions
  • the device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device.
  • Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

Abstract

一种基于栅格时延预测的卫星传输优化方法，首先，根据卫星星座的参数，基于地理栅格划分网格优先级，确定用户与卫星的关联和切换方式；其次，根据任务的卸载位置，建立任务的时延模型；然后，根据随机网络演算推导的栅格时延边界设定任务限制时延，作为卫星的栅格资源约束条件，并设定目标优化函数；之后，考虑多个任务卸载至同一节点竞争资源的情况，基于栅格时延保障研究公平的资源分配方案；最后，在确定的资源分配方案下，研究用户任务的卸载和调度决策。

Description

一种基于栅格时延预测的卫星传输优化方法 技术领域
本发明属于卫星传输任务卸载和资源分配技术领域，尤其涉及一种基于栅格时延预测的卫星传输优化方法。
背景技术
由于单颗卫星的覆盖范围可以达到数万平方公里量级，其覆盖区域将跨越多种地理环境。同时，由于卫星相对地面高速运动，其服务范围内的业务特征呈现快速时变的特性，从而导致全球范围内的卫星物联网业务量分布存在较明显的时空不均匀性。
关于卫星网络业务建模与分析领域的相关研究大多未考虑地理位置的问题，而研究与地理位置信息相关的业务模型是解决卫星物联网海量节点随机接入业务碰撞的重要基础。国内外针对卫星物联网架构的传输优化策略研究己经非常丰富，但针对基于移动边缘计算的卫星物联网架构，大多研究未考虑业务QoS服务质量的保障问题，尚缺乏基于时延预测的传输优化策略。
发明内容
针对现有技术的不足，本发明的目的在于提供一种基于栅格时延预测的卫星传输优化方法，针对卫星运动导致的时变特性、卫星网络业务到达的突发性、系统服务随机性问题，采用基于地理栅格和随机网络演算的卫星物联网业务建模分析方法，推导栅格时延边界作为Qos服务质量保障，并基于栅格统计时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题，为解决该问题，将其解耦为基于栅格时延保障的公平资源分配和任务卸载与调度决策两个子问题，分别采用对偶上升法和深度Q网络(DQN)对两个子问题进行求解，以提高卫星网络的资源利用率。
本发明为了解决以上问题采用了以下技术方案：
第一方面，本发明提供了一种卫星传输优化方法，包括：
采用基于地理栅格和随机网络演算的卫星物联网业务建模分析方法，推导出栅格时延边界，基于所述栅格时延边界设定栅格时延限制值，作为栅格时延保障；
基于栅格时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题P1；
将问题P1解耦为基于栅格时延保障的公平资源分配子问题和任务卸载与调度决策子问题；
采用对偶上升法对公平通信与计算资源分配子问题进行求解，得到公平的通信资源分配和计算资源的分配；
基于所述通信资源分配和计算资源分配，采用深度Q网络(DQN)对任务卸载与调度决策子问题进行求解，得到用户当前任务的卸载决策和调度决策。
在一些实施例中，基于栅格时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题P1，包括：
Figure PCTCN2022112499-appb-000001
其中，w为l个时隙内完成并返回结果的用户任务总数；当栅格i范围内的用户m的任务k在基于SNC推导而设置的限定时延
Figure PCTCN2022112499-appb-000002
内完成并返回结果，即
Figure PCTCN2022112499-appb-000003
时，
Figure PCTCN2022112499-appb-000004
否则为0；
Figure PCTCN2022112499-appb-000005
为第l个时隙在本地处理的任务的集合，
Figure PCTCN2022112499-appb-000006
表示任务在本地处理；
Figure PCTCN2022112499-appb-000007
为第l个时隙用户任务卸载的集合，
Figure PCTCN2022112499-appb-000008
Figure PCTCN2022112499-appb-000009
表示用户m的任务卸载至卫星n处理；
Figure PCTCN2022112499-appb-000010
为第l个时隙用户任务的调度集合，
Figure PCTCN2022112499-appb-000011
表示从卫星的传输或处理队列中调度用户m的任务，否则不调度任务；C l＝{C 1,l,C 2,l,...,C N,l},
Figure PCTCN2022112499-appb-000012
为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
Figure PCTCN2022112499-appb-000013
为第l个时隙卫星给各用户分配的计算资源；
Figure PCTCN2022112499-appb-000014
和
Figure PCTCN2022112499-appb-000015
分别为第l个时隙卫星n可分配的通信资源和计算资源；
Figure PCTCN2022112499-appb-000016
和
Figure PCTCN2022112499-appb-000017
分别为第l个时隙从卫星n的传输和处理队列中调度的任务集合；
s.t.表示约束条件，约束条件式(18a)保证第l个时隙卫星给用户m分配的通信资源
Figure PCTCN2022112499-appb-000018
不应大于第l个时隙卫星n可分配的通信资源
Figure PCTCN2022112499-appb-000019
约束条件式(18b)保证第l个时隙卫星给 用户m分配的计算资源
Figure PCTCN2022112499-appb-000020
不应大于第l个时隙卫星n可分配的计算资源
Figure PCTCN2022112499-appb-000021
约束条件式(18c)保证用户m的链路时延
Figure PCTCN2022112499-appb-000022
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000023
在一些实施例中，在第l个时隙，设卸载决策
Figure PCTCN2022112499-appb-000024
和调度决策
Figure PCTCN2022112499-appb-000025
已知，卫星的公平通信与计算资源分配子问题表述为：
Figure PCTCN2022112499-appb-000026
Figure PCTCN2022112499-appb-000027
其中，
Figure PCTCN2022112499-appb-000028
为栅格i范围的用户m的任务从卫星n的传输队列中调度并分配通信资源后产生的传输时延；
Figure PCTCN2022112499-appb-000029
为栅格i范围的用户m的任务从卫星n的处理队列中调度并分配计算资源后产生的处理时延；
Figure PCTCN2022112499-appb-000030
为基于SNC设定的栅格i的限定时延；C l＝{C 1,l,C 2,l,...,C N,l},
Figure PCTCN2022112499-appb-000031
为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
Figure PCTCN2022112499-appb-000032
为第l个时隙卫星给各用户分配的计算资源；
Figure PCTCN2022112499-appb-000033
和
Figure PCTCN2022112499-appb-000034
分别为第l个时隙卫星n可分配的通信资源和计算资源；
s.t.表示约束条件，约束条件式(22a)保证第l个时隙卫星n给用户m分配的通信资源
Figure PCTCN2022112499-appb-000035
不应大于第l个时隙卫星n可分配的通信资源
Figure PCTCN2022112499-appb-000036
约束条件式(22b)保证用户m的链路时延
Figure PCTCN2022112499-appb-000037
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000038
约束条件式(23a)保证第l个时隙卫星给用户m分配的计算资源
Figure PCTCN2022112499-appb-000039
不应大于第l个时隙卫星n可分配的计算资源
Figure PCTCN2022112499-appb-000040
约束条件 式(23b)保证用户m的链路时延
Figure PCTCN2022112499-appb-000041
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000042
在一些实施例中，采用对偶上升法对公平通信与计算资源分配子问题进行求解，包括：
采用对偶上升法对通信资源分配子问题进行求解：
引入辅助变量χ，传输时延
Figure PCTCN2022112499-appb-000043
其中T m为用户m的任务k的大小；
公式(22)转换为：
Figure PCTCN2022112499-appb-000044
构造拉格朗日函数A：
Figure PCTCN2022112499-appb-000045
其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子；
则A的对偶函数为：
Figure PCTCN2022112499-appb-000046
其中，
Figure PCTCN2022112499-appb-000047
D的最大值即为公式(22)所求的最小值；
通过交替迭代辅助变量χ和拉格朗日乘子μ m、ν、η，得到公平的通信资源分配
Figure PCTCN2022112499-appb-000048
同理，采用对偶上升法对计算资源分配子问题进行求解：
引入辅助变量χ，处理时延
Figure PCTCN2022112499-appb-000049
其中T m为用户m的任务k的大小；
公式(23)转换为：
Figure PCTCN2022112499-appb-000050
Figure PCTCN2022112499-appb-000051
构造拉格朗日函数A：
Figure PCTCN2022112499-appb-000052
其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子；
则A的对偶函数为：
Figure PCTCN2022112499-appb-000053
其中，
Figure PCTCN2022112499-appb-000054
D的最大值即为公式(23)所求的最小值；
通过交替迭代辅助变量χ和拉格朗日乘子μ m、ν、η，得到公平的计算资源分配
Figure PCTCN2022112499-appb-000055
在一些实施例中，采用深度Q网络对任务卸载与调度决策子问题进行求解，包括：
将物联网终端的信息：P U(l)、 X U,l、T l，卫星的信息：P s(l)、 Β l、 C l、 X l，队列的信息：Q trans,l、Q pro,l输入预训练好的深度Q网络；
得到输出的任务的卸载决策
Figure PCTCN2022112499-appb-000056
和调度决策
Figure PCTCN2022112499-appb-000057
定义第l个时隙的状态h l＝{P U(l),P s(l),T l, Β l, X U,l, C l, X l,Q trans,l,Q pro,lQ local,l}，P U(l)，P s(l)分别为第l个时隙用户和卫星的位置；T l＝{T 1,l,T 2,l,...,T M,l}为第l个时隙等待或正在调度的用户任务的大小； Β l＝{ Β 1,l, Β 2,l,..., Β M,l}为第l个时隙用户关联卫星的情况， Β m,l∈{1,2...,N}； X U,l＝{ X 1,l, X 2,l,..., X M,l}， X m,l∈{0,1}表示第l个时隙用户本地的计算资源是否被占用，即是否有任务正在本地处理； C l＝{ C 1,l, C 2,l,..., C N,l}为第l个时隙卫星被占用的通信资源； X l＝{ X 1,l, X 2,l,..., X N,l}为第l个时隙卫星被占用的计算资源；
Figure PCTCN2022112499-appb-000058
和
Figure PCTCN2022112499-appb-000059
分别为第l个时隙卫星的传输队列和处理队列中等待调度的任务总量；
Figure PCTCN2022112499-appb-000060
为第l个时隙本地用户任务到达队列中等待调度的任 务总量；
从任务卸载和调度决策的角度定义第l个时隙的动作
Figure PCTCN2022112499-appb-000061
Figure PCTCN2022112499-appb-000062
其中，E off∈{1,2...,Z}为任务的卸载决策，E off＝0表示任务在本地处理，E off＝Z表示任务卸载至卫星Z处理，E exe∈{0,1}表示是否从卫星的传输或处理队列中调度任务；通过动作a l即可得到用户当前任务的卸载决策
Figure PCTCN2022112499-appb-000063
和调度决策
Figure PCTCN2022112499-appb-000064
在一些实施例中，栅格时延边界表示为：
Figure PCTCN2022112499-appb-000065
其中D(n)为时延函数，P(D(n)＞x)表示时延超过某个值x的概率，β(n)为随机服务曲线，β(n)是关于n的线性函数，即β(n)＝ρ β(θ)n，ρ β是随机服务曲线β(n)的斜率函数，θ、θ 1均为大于0的自由参数。
第二方面，本发明提供了一种卫星传输优化装置，包括处理器及存储介质；
所述存储介质用于存储指令；
所述处理器用于根据所述指令进行操作以执行根据第一方面所述方法的步骤。
第三方面，本发明提供了一种存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现第一方面所述方法的步骤。
与现有技术相比，本发明具有以下技术效果：本发明根据随机网络演算推导的栅格时延边界设定任务限制时延，作为卫星的栅格资源约束条件，能够在保障栅格业务QoS、用户公平性的基础上实现动态网络环境中路由关键性能指标(KPI)的快速、准确、低开销评估与推断，提高卫星网络的资源利用率。
附图说明
图1是本发明实施例中基于栅格时延预测的卫星传输优化方法的流程图。
图2是本发明实施例中栅格业务建模的流程图。
图3是本发明实施例中融合星地协同网络架构图。
图4是本发明实施例中联合算法流程图。
具体实施方式
下面结合附图对本发明作进一步描述，以下实施例仅用于更加清楚地说明本发明的技术 方案。
在本发明的描述中，若干的含义是一个以上，多个的含义是两个以上，大于、小于、超过等理解为不包括本数，以上、以下、以内等理解为包括本数。如果有描述到第一、第二只是用于区分技术特征为目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量或者隐含指明所指示的技术特征的先后关系。
本发明的描述中，参考术语“一个实施例”、“一些实施例”、“示意性实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本发明的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不一定指的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任何的一个或多个实施例或示例中以合适的方式结合。
实施例1
一种卫星传输优化方法，包括：
采用基于地理栅格和随机网络演算的卫星物联网业务建模分析方法，推导出栅格时延边界，将栅格时延边界作为服务质量保障；基于所述栅格时延边界设定栅格时延限制值，作为栅格时延保障；(将栅格i范围内的用户m的任务k在基于SNC随机网络演算推导而设置的限定时延表示为
Figure PCTCN2022112499-appb-000066
基于栅格时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题P1；
将问题P1解耦为基于栅格时延保障的公平资源分配子问题和任务卸载与调度决策子问题；
采用对偶上升法对公平通信与计算资源分配子问题进行求解，得到公平的通信资源分配和计算资源的分配；
基于所述通信资源分配和计算资源分配，采用深度Q网络(DQN)对任务卸载与调度决策子问题进行求解，得到用户当前任务的卸载决策和调度决策。
在一些实施例中，如图1所示，一种基于栅格时延预测的卫星传输优化方法，包括：
步骤一.基于地理栅格和随机网络演算的网络业务建模与分析：
与地面物联网相对静态的拓扑结构不同，由于卫星节点的高动态特性，终端每次产生的业务都可能经由不同的卫星节点汇聚。其中卫星星下点运动轨迹可由卫星星历推算得出，本方案采用随机网络演算理论结合地理栅格的方法分析卫星网络的排队性能，在选取卫星网络 的业务到达和信道服务模型时，需充分考虑卫星网络业务到达的随机性和突发性，以及多址接入协议引入的系统服务随机性。研究思路如下：
如图2所示，首先，分析全球物联网终端部署，划分地理栅格；然后，基于随机网络演算进行以栅格为单位的业务建模，通过分析网络中业务到达特性和系统服务特性，选取合适的到达模型和服务模型，推导相应的随机到达曲线和随机服务曲线；最后推导各栅格网络性能边界。
(1)地理栅格划分
由于卫星的广覆盖特性，在卫星波束的覆盖区域中势必涵盖多样的地理环境，因而覆盖区内的业务将呈现多样性。因此为了确定覆盖区域内的业务特征，必须要系统性地分析覆盖区域的地理环境情况。首先将地球表面展开看成一二维平面，并按一定的经纬度间隔将地球表面划分为许多栅格，由于划分后的栅格中可能会包含多种地理环境，所以某栅格的设备部署密度由多种地理环境共同决定，定义栅格i的设备部署密度为：
Figure PCTCN2022112499-appb-000067
其中S i表示栅格i的面积，n表示该栅格内包含的地理环境类型数量，ρ j表示在地理环境类型j对应的设备部署密度，S i,j表示栅格i内地理环境类型j所占的面积。S i的计算方式为：
Figure PCTCN2022112499-appb-000068
其中，
Figure PCTCN2022112499-appb-000069
和
Figure PCTCN2022112499-appb-000070
为栅格i纬度的起止值，μ是该栅格的经度范围，a e是地球赤道圆的半径，e 1为第一离心率。
卫星波束在地面的投影将直接决定其覆盖范围，从而决定其覆盖区域内的栅格数目。由于物联网业务汇聚于卫星层面，为了简化覆盖面积的计算，可将卫星的全部点波束看成一个合成波束，则其在地面的投影可近似看成为以星下点为圆心的圆周所围区域，如图所示，其圆心角为：
Figure PCTCN2022112499-appb-000071
其中，R e为地球半径，H为轨道高度，α为最小通信仰角。
(2)业务到达模型
现有研究成果验证了卫星网络业务具有自相似性和突发特性。随机网络演算理论是一种新型排队理论分析方法，该理论并不限定网络中业务到达和服务服某种特定分布，而是用包络的方法对网络中的业务到达和信道服务建模，并在模型中引入违反概率，可精确描述卫星网络中业务产生的随机性、突发性以及信道服务的随机性。
因此，为了精确描述卫星网络业务的突发特性，并在模型准确性和理论分析的简化之间取得折中，采用马尔可夫调制开关过程(Markov-modulated ON-OFF，MMOO)对卫星网络的业务到达进行表征，该模型具有较大可调节性，可适用于不同程度的突发业务。
对于业务到达过程A(n)，我们将其用两状态马尔科夫链来表征，其状态空间S由0和1组成。状态1表示发送节点处于ON状态，以固定速率h发送数据包；状态0表发送节点处于OFF状态，此时无数据包产生。从0状态转换到1状态的概率为μ，而从1转换为0的概率为λ。
卫星网络业务到达过程A(n)具有v.b.c.随机到达曲线，用马尔科夫调制的开关过程表征的业务到达A(n),
Figure PCTCN2022112499-appb-000072
有
Figure PCTCN2022112499-appb-000073
因此
Figure PCTCN2022112499-appb-000074
令
Figure PCTCN2022112499-appb-000075
卫星网络业务到达过程A(n)具有v.b.c.随机到达曲线，即：A(n)～ sac<f,a>，其中
Figure PCTCN2022112499-appb-000076
有
α(n)＝ρ α(θ)n               (7)
Figure PCTCN2022112499-appb-000077
Figure PCTCN2022112499-appb-000078
由此可见，α(n)是关于n的线性函数，ρ α(θ)是它的斜率，f(x)是α(n)的边界函数。
(3)信道服务模型
在卫星网络的场景中，服务随机性体现在MAC层多用户通过随机接入协议竞争信道服务带来的随机性。目前卫星网络中多采用时隙Aloha协议实现信道预约，因此本方案采用两状态马尔科夫链对时隙Aloha协议的工作原理进行表征。在状态1时，链路可提供速率C来传输数据，在状态0时，链路无法为数据包提供服务。当信道从状态0转换到状态1时，可实现一次数据包成功传输，相应的状态转移概率即为一次成功传输数据的概率p s。当信道从状态1转换为状态0，数据包发生碰撞导致传输失败，相应的状态转移概率为1-p s。
卫星网络的随机服务过程S(n)有随机服务曲线S(n)～ s-ssc<g,β>，其中：
Figure PCTCN2022112499-appb-000079
g(x)＝e -θx             (11)
Figure PCTCN2022112499-appb-000080
由此可见，β(n)是关于n的线性函数，ρ β(θ)是它的斜率，g(x)是β(n)的边界函数。
(4)栅格时延边界分析
卫星网络时延边界表达式为：
Figure PCTCN2022112499-appb-000081
其中D(n)为时延函数，P(D(n)＞x)表示时延超过某个值x的概率，β(n)为随机服务曲线，β(n)是关于n的线性函数，即β(n)＝ρ β(θ)n，ρ β是随机服务曲线β(n)的斜率函数，θ、θ 1均为大于0的自由参数。
证明：如果系统到达过程与服务过程相互独立，那么系统时延边界的计算可转化为如下：
Figure PCTCN2022112499-appb-000082
根据函数最大水平距离的定义：
h(α+x,β)＝sup m≥0{inf{τ≥0:α(m)+x≤β(m+τ)}}        (15)
可以理解为：求出τ的最小值，使得α(s)+x≤β(s+τ)在s≥0时恒成立。因此，当s＝0 时，x＝β(τ)＝ρ βτ。将x的值带入时延求解公式(14)可得：
Figure PCTCN2022112499-appb-000083
将
Figure PCTCN2022112499-appb-000084
g(x)＝e -θx带入公式(16)，可得：
Figure PCTCN2022112499-appb-000085
结论得证。
(5)基于上述四步原理划分地理栅格，基于随机网络演算建立栅格业务到达模型和服务模型，并推导出栅格性能边界，再基于边界设定一个栅格时延限制值，将栅格i范围内的用户m的任务k在基于SNC推导而设置的限定时延表示为
Figure PCTCN2022112499-appb-000086
该栅格业务建模与分析方法可以有效地为卫星网络资源分配设计提供Qos保障。
步骤二.融合MEC场景下基于栅格统计时延保障的任务调度与资源分配策略
现有研究大多只关注卫星网络资源管理的单个或两个方面，本方案联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题，为解决该问题，将其解耦为基于栅格时延保障的公平资源分配和任务卸载与调度决策两个子问题，分别采用对偶上升法和深度Q网络(DQN)进行求解。具体过程如下：
1)我们将具有四个耦合因素的复杂问题解耦为两个子问题，首先是具有栅格时延保障以及固定卸载决策的计算和通信资源分配问题，其次是具有动态约束的卸载与调度决策问题。
2)为联合优化具有栅格时延保障以及固定卸载决策的卫星计算和通信资源，采用对偶上升法求解最优的通信和计算资源分配。然后，将资源分配子问题的结果输入到卸载决策问题中，将具有动态约束的卸载决策问题制定为马尔可夫决策过程(MDP)，并使用深度Q网络(DQN)来增加卸载决策的长期回报，提高任务的完成率。
研究思路：首先，根据卫星星座的参数，基于地理栅格划分网格优先级，确定用户与卫星的关联和切换方式；其次，并根据任务的卸载位置，建立任务的时延模型；然后，根据随机网络演算推导的栅格时延边界设定任务限制时延，作为卫星的栅格资源约束条件，并设定目标优化函数；之后，考虑多个任务卸载至同一节点竞争资源的情况，基于栅格时延保障研究公平的资源分配方案；最后，在确定的资源分配方式下，研究用户任务的卸载和调度决策，分析其在提高系统任务完成率方面的表现。
场景介绍：如图3所示，在融合MEC的星地协同网络架构的应用场景下，地面上有M个物联网终端用户，M个用户都会有持续到达的任务，空中有N颗卫星组成的卫星星座，卫星可通过星间无线链路与其他卫星通信，以实现对用户任务的星间协作处理。用户的任务可在本地处理或卸载至卫星处理，本地同时只能处理一个任务，星上可并行处理多个任务。卸载至卫星的任务可在与用户关联的接入星处理并在处理完成后将结果返回用户。当接入星的资源难以满足用户需求时，可以通过星间无线链路将任务卸载至周围其他卫星进行处理，以额外10ms左右的传播时延换取更低的处理时延，以达到星间协作处理的优化效果。
(1)问题建模及分析
在本发明的研究场景中，通过星间无线链路实现卫星网络对地面用户任务的协作处理，为了尽量保证用户任务在限定的时延内完成，需要对任务卸载决策、调度决策、通信和计算资源分配联合进行优化。因此，本发明的最优化问题可表述为，
Figure PCTCN2022112499-appb-000087
其中，w为l个时隙内完成并返回结果的用户任务总数；当栅格i范围内的用户m的任务k在基于SNC推导而设置的限定时延
Figure PCTCN2022112499-appb-000088
内完成并返回结果，即
Figure PCTCN2022112499-appb-000089
时，
Figure PCTCN2022112499-appb-000090
否则为0；
Figure PCTCN2022112499-appb-000091
为第l个时隙在本地处理的任务的集合，
Figure PCTCN2022112499-appb-000092
表示任务在本地处理；
Figure PCTCN2022112499-appb-000093
为第l个时隙用户任务卸载的集合，
Figure PCTCN2022112499-appb-000094
Figure PCTCN2022112499-appb-000095
表示用户m的任务卸载至卫星n处理；
Figure PCTCN2022112499-appb-000096
为第l个时隙用户任务的调度集合，
Figure PCTCN2022112499-appb-000097
表示从卫星的传输或处理队列中调度用户m的任务，否则不调度任务；C l＝{C 1,l,C 2,l,...,C N,l},
Figure PCTCN2022112499-appb-000098
为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
Figure PCTCN2022112499-appb-000099
为第l个时隙卫星给各用户分配的计算资源；
Figure PCTCN2022112499-appb-000100
和
Figure PCTCN2022112499-appb-000101
分别为第l个时隙卫星n可分配的通信和计算资源；
Figure PCTCN2022112499-appb-000102
和
Figure PCTCN2022112499-appb-000103
分别为第l个时隙从卫星n的传输和处理队列中调度的任务集合。
约束条件式(18a)保证第l个时隙卫星给用户m分配的通信资源
Figure PCTCN2022112499-appb-000104
不应大于第l个时隙卫星n可分配的通信资源
Figure PCTCN2022112499-appb-000105
约束条件式(18b)保证第l个时隙卫星给用户m分配的计算资源
Figure PCTCN2022112499-appb-000106
不应大于第l个时隙卫星n可分配的计算资源
Figure PCTCN2022112499-appb-000107
约束条件式(18c)保证用户m的链路时延
Figure PCTCN2022112499-appb-000108
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000109
(2)算法流程
完整的算法流程如图4所示，首先基于地理栅格划分和随机网络建模推导栅格业务限制时延，然后在每个调度时隙，判断是否因为卫星的运动发生切星，之后由基于DQN的任务卸载与调度算法，对持续到达的任务进行卸载和调度，从而对于在同一颗卫星的传输/处理队列中同时调度的任务，由基于对偶上升法的公平资源分配算法为其分配通信与计算资源，即在己知任务卸载与调度决策的前提下，将通信与计算资源的分配问题建模为最大－最小公平性问题进行求解，以提高系统任务完成率。在系统运行期间重复上述步骤。
(3)时延模型
根据任务的卸载位置不同，用户m的任务k产生的时延
Figure PCTCN2022112499-appb-000110
共有三种情况。
①如果任务在本地处理，其时延为，
Figure PCTCN2022112499-appb-000111
其中，τ wait表示由于本地计算资源被正在处理的任务占用而产生的等待时延；
τ process＝T m/X m，X m为用户m的计算资源，T m为用户m的任务k的大小。
②如果任务卸载至与其关联的接入星n a处理，其时延为，
Figure PCTCN2022112499-appb-000112
其中，τ off为任务从用户卸载到接入星n a的时延，包括传输等待时延τ wait、传输时延τ trans、传播时延τ prop三部分，
Figure PCTCN2022112499-appb-000113
为接入星n a给用户m分配的通信资源，
Figure PCTCN2022112499-appb-000114
c为光速，为用户m到接入星n a的传播距离；τ process由等待处理时延和处理时延两部分组成，
Figure PCTCN2022112499-appb-000115
为接入星n a给用户m分配的计算资源，需要注意的是，为了避免资源浪费，任务传输完成后，才可进入处理队列分配计算资源，此处
Figure PCTCN2022112499-appb-000116
和
Figure PCTCN2022112499-appb-000117
的分配是独立且有先后顺序的，不一定在同一时隙完成；τ return为结果返回时延，由于返回的处理 结果数据量很小，因此忽略传输时延，只考虑返回的传播时延。如果处理结果可在接入星结束覆盖前返回用户，
Figure PCTCN2022112499-appb-000118
否则，
Figure PCTCN2022112499-appb-000119
而为接入星n a到切换星n h的最短路由距离，可由Dijkstra算法求得，
Figure PCTCN2022112499-appb-000120
为切换星n h到用户m的传播距离。
③如果任务通过与其关联的接入星n a卸载至处理星n p处理，其时延为
Figure PCTCN2022112499-appb-000121
其中，τ off为任务从用户m卸载到接入星n a的时延，计算方式同(20)一致；τ ISL为任务从接入星n a通过星间链路卸载到处理星n p的时延，考虑到星间充足的通信资源，星间链路的传输时延可以忽略不计，因此
Figure PCTCN2022112499-appb-000122
为接入星n a到处理星n p的星间路由距离(单跳或多跳)，由Dijkstra算法求得；τ process包括任务在处理星的处理队列的等待时延τ w_pro和处理时延τ process两部分；τ return为处理结果从处理星n p返回用户m的传播时延，首先处理结果会先从处理星n p返回接入星n a，再由接入星返回给用户，如果处理结果可在接入星结束覆盖前返回用户，
Figure PCTCN2022112499-appb-000123
否则，
Figure PCTCN2022112499-appb-000124
为处理星n p到切换星n h的星间路由距离。
(4)卫星覆盖与切换策略
依据前文所述卫星波束在地面的投影将直接决定其覆盖范围，从而决定其覆盖区域内的栅格数目，由于系统需对全球提供无缝覆盖，在系统设计时卫星覆盖区域会出现部分重叠。在重叠区域内，终端将根据距离最近原则优先选择卫星节点进行传输。
卫星覆盖区域按照距星下点的距离被划分为若干优先级网格，在卫星重叠覆盖区域内，处在同一个地理栅格中的所有物联网设备节点实时选择优先级最高的卫星进行汇聚。当通信仰角小于最小覆盖仰角α min时，表示该卫星即将结束覆盖终端，重新根据该优先级准则选择优先级更高的卫星进行切换。如果任务处理结果未能在接入星结束覆盖前返回，则通过切换的卫星将结果返回。
以某一个时刻为例，某一颗卫星相对于地理网格的优先级按照如下的步骤进行计算：
步骤1、取出卫星星下点所处的经纬度坐标，确定其处于地理网格的网格索引。
步骤2、将卫星星下点所处的网格的优先级设置为1，为其周围网格设置相应的优先级，距离此网格越近，表示与卫星的距离越小，则优先级越高。
步骤3、网格取出相对于自身优先级最高的卫星编号进行接入。
(5)基于对偶上升法的公平资源分配算法
为了求解任务卸载与调度决策的子问题，需要先求得资源分配子问题的解。在己知任务卸载与调度决策的前提下，将通信与计算资源的分配问题建模为最大－最小公平性问题进行求解，以最小化任务的最大时延，提高系统整体的任务完成率。详细的求解推导过程如下文所示。
在第l个时隙，如果卸载决策
Figure PCTCN2022112499-appb-000125
和调度决策
Figure PCTCN2022112499-appb-000126
已知，卫星的公平资源分配问题可表述为，
Figure PCTCN2022112499-appb-000127
Figure PCTCN2022112499-appb-000128
其中，
Figure PCTCN2022112499-appb-000129
为栅格i范围的用户m的任务从卫星n的传输队列中调度并分配通信资源后产生的传输时延；
Figure PCTCN2022112499-appb-000130
为栅格i范围的用户m的任务从卫星n的处理队列中调度并分配计算资源后产生的处理时延；
Figure PCTCN2022112499-appb-000131
为基于SNC设定的栅格i的限定时延；C l＝{C 1,l,C 2,l,...,C N,l},
Figure PCTCN2022112499-appb-000132
为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
Figure PCTCN2022112499-appb-000133
为第l个时隙卫星给各用户分配的计算资源；
Figure PCTCN2022112499-appb-000134
和
Figure PCTCN2022112499-appb-000135
分别为第l个时隙卫星n可分配的通信和计算资源。
约束条件式(22a)保证第l个时隙卫星给用户m分配的通信资源
Figure PCTCN2022112499-appb-000136
不应大于第l个时隙卫星n可分配的通信资源
Figure PCTCN2022112499-appb-000137
约束条件式(22b)保证用户m的链路时延
Figure PCTCN2022112499-appb-000138
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000139
同理，约束条件式(23a)保证第l个时隙卫星给用户m分配的计算资源
Figure PCTCN2022112499-appb-000140
不应大于第l个时隙卫星n可分配的计算资源
Figure PCTCN2022112499-appb-000141
约束条件式(23b)保证 用户m的链路时延
Figure PCTCN2022112499-appb-000142
不应大于基于SNC设定的栅格i的限定时延
Figure PCTCN2022112499-appb-000143
由于最优化问题(22)和(23)是两个凸问题，可由对偶上升法(Dual Ascent)进行求解。
首先，采用对偶上升法对通信资源分配子问题进行求解：
引入辅助变量χ，由前文可知传输时延
Figure PCTCN2022112499-appb-000144
则公式(22)可转换为，
Figure PCTCN2022112499-appb-000145
其中，引入辅助变量χ后，优化模型约束条件由(22a-b)转换为(24a-c)。
其次，构造拉格朗日函数A：
Figure PCTCN2022112499-appb-000146
其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子。
则A的对偶函数为：
Figure PCTCN2022112499-appb-000147
其中，
Figure PCTCN2022112499-appb-000148
由于公式(22)为凸问题，则D的最大值即为公式(22)所求的最小值。求解过程如下所示。
Figure PCTCN2022112499-appb-000149
Figure PCTCN2022112499-appb-000150
如果μ m、ν、η的更新值连续100次迭代小于0.001，则认为其己收敛。通过交替迭代自变量和拉格朗日乘子，可以得到公平的通信资源分配
Figure PCTCN2022112499-appb-000151
同理，采用对偶上升法对计算资源分配子问题进行求解，包括：
引入辅助变量χ，处理时延
Figure PCTCN2022112499-appb-000152
其中T m为用户m的任务k的大小
公式(23)转换为：
Figure PCTCN2022112499-appb-000153
构造拉格朗日函数A：
Figure PCTCN2022112499-appb-000154
其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子；
则A的对偶函数为：
Figure PCTCN2022112499-appb-000155
其中，
Figure PCTCN2022112499-appb-000156
D的最大值即为公式(23)所求的最小值；
通过交替迭代辅助变量χ和拉格朗日乘子μ m、ν、η，得到公平的计算资源分配
Figure PCTCN2022112499-appb-000157
求解过程如下所示。
Figure PCTCN2022112499-appb-000158
Figure PCTCN2022112499-appb-000159
(6)基于DQN的任务卸载与调度决策算法
由前文可得到每个时隙最公平的通信和计算资源分配。然而，卸载与调度的联合决策仍然是一个非凸的动态规划问题。因此采用DQN算法，解决持续到达的多批任务在长时间段内耦合的卸载、调度、资源分配问题。此问题具体的MDP表示如下：
1)Stata(H)：
定义第l个时隙的状态h l＝{P U(l),P s(l),T l, Β l, X U,l, C l, X l,Q trans,l,Q pro,lQ local,l}，P U(l)，P s(l)分别为第l个时隙用户和卫星的位置；T l＝{T 1,l,T 2,l,...,T M,l}为第l个时隙等待或正在调度的用户任务的大小； Β l＝{ Β 1,l, Β 2,l,..., Β M,l}为第l个时隙用户关联卫星的情况， Β m,l∈{1,2...,N}； X U,l＝{ X 1,l, X 2,l,..., X M,l}， X m,l∈{0,1}表示第l个时隙用户本地的计算资源是否被占用，即是否有任务正在本地处理； C l＝{ C 1,l, C 2,l,..., C N,l}为第l个时隙卫星被占用的通信资源； X l＝{ X 1,l, X 2,l,..., X N,l}为第l个时隙卫星被占用的计算资源；
Figure PCTCN2022112499-appb-000160
和
Figure PCTCN2022112499-appb-000161
分别为第l个时隙卫星的传输队列和处理队列中等待调度的任务总量；
Figure PCTCN2022112499-appb-000162
为第l个时隙本地用户任务到达队列中等待调度的任务总量。
2)Action(A)
每个时隙的动作空间应包括用户当前任务的卸载、调度决策和卫星通信、计算资源的分配。由于通信资源分配C l和计算资源分配X l可由算法1-1和算法1-2得到，因此，从任务卸载和调度决策的角度定义第l个时隙的动作
Figure PCTCN2022112499-appb-000163
其中， E off∈{1,2...,Z}为任务的卸载决策，E off＝0表示任务在本地处理，E off＝Z表示任务卸载至卫星Z处理，由于整个星座的卫星个数N通常在几百到几千，大部分离用户距离很远，因此，仅考虑离用户最近的Z颗卫星进行卸载；E exe∈{0,1}表示是否从卫星的传输或处理队列中调度任务。通过a l即可得到用户当前任务的卸载决策
Figure PCTCN2022112499-appb-000164
和调度决策
Figure PCTCN2022112499-appb-000165
3)Transition Probability(P)
在本方案所研究的场景中，动作空间和状态空间都非常庞大，且状态空间中的部分量为连续变量，难以得到精确的状态转移概率。因此，本方案选择了model-free的深度强化学习DQN架构。
4)Reward(R)
为了最大化用户任务的完成率，定义第l个时隙状态h l下选择动作a l的回报为：
Figure PCTCN2022112499-appb-000166
其中，
Figure PCTCN2022112499-appb-000167
为第l个时隙调度任务k而产生的传输或处理时延；R P是一个正的常数，使
Figure PCTCN2022112499-appb-000168
d为第l个时隙在限制时延内完成的任务个数；R d为任务在限制时延内完成的额外回报。
采用深度Q网络对任务卸载与调度决策子问题进行求解，包括：
将物联网终端的信息：P U(l)、 X U,l、T l，卫星的信息：P s(l)、 Β l、 C l、 X l，队列的信息：Q trans,l、Q pro,l输入预训练好的深度Q网络；
得到输出的任务的卸载决策
Figure PCTCN2022112499-appb-000169
和调度决策
Figure PCTCN2022112499-appb-000170
深度Q网络的训练过程如算法2所示，其中，γ、ε、ζ、δ为训练寻优过程中涉及的网络参数：γ是动作价值函数的折扣因子；ε为贪心策略选择动作概率(贪心策略，就是小概率选择随机动作，大概率选择最优动作)，即以ε的概率随机选择动作，以1-ε的概率选择最优动作；ζ为经验回放池；δ用于计算训练损失函数，其训练目标为使损失函数最小化。
每个时隙进行一次任务调度，每个时隙为训练过程的一步。首先以参数γ、ε、ζ初始化网络和经验池。其次在每个时隙l的状态h l下，根据ε贪心策略选取动作，即以ε的概率随机 选择动作，以1-ε的概率选择最优动作。然后输入根据算法1-1和算法1-2，计算得到的卫星通信资源分配 C l和计算资源的分配 X l。接着更新系统环境至下一状态h l+1，计算本时隙的状态h l下选择动作a l的回报R(h l,a l)，并将h l和a l保存至经验池中。最后计算损失函数的值，每隔固定的步数，从主网络中复制参数到目标网络。重复上述步骤直至损失函数收敛于0。
Figure PCTCN2022112499-appb-000171
实施例2
第二方面，本实施例提供了一种卫星传输优化装置，包括处理器及存储介质；
所述存储介质用于存储指令；
所述处理器用于根据所述指令进行操作以执行根据实施例1所述方法的步骤。
实施例3
第三方面，本实施例提供了一种存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现实施例1所述方法的步骤。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
本领域内的技术人员应该明白，本申请的实施例可提供为方法或计算机程序产品。以上实施例仅用于说明本发明所提出的方法而并非限制本方法，尽管上文通过实施例对本专利方法进行了详细的说明，所述领域的技术人员应当理解，仅对本发明具体实施方法进行同等替换而其本质并未改变的方案应包含在本发明的权利要求保护范围之内。

Claims (9)

  1. 一种卫星传输优化方法，其特征在于，包括：
    采用基于地理栅格和随机网络演算的卫星物联网业务建模分析方法，推导出栅格时延边界，基于所述栅格时延边界设定栅格时延限制值，作为栅格时延保障；
    基于栅格时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题P1；
    将问题P1解耦为基于栅格时延保障的公平资源分配子问题和任务卸载与调度决策子问题；
    采用对偶上升法对公平通信与计算资源分配子问题进行求解，得到公平的通信资源分配和计算资源的分配；
    基于所述通信资源分配和计算资源分配，采用深度Q网络对任务卸载与调度决策子问题进行求解，得到用户当前任务的卸载决策和调度决策。
  2. 根据权利要求1所述的卫星传输优化方法，其特征在于，基于栅格时延保障，联合考虑任务的卸载决策、调度决策、通信和计算资源的分配方式，将优化问题表述为混合整数非线性规划问题P1，包括：
    Figure PCTCN2022112499-appb-100001
    其中，w为l个时隙内完成并返回结果的用户任务总数；当栅格i范围内的用 户m的任务k在基于SNC推导而设置的限定时延
    Figure PCTCN2022112499-appb-100002
    内完成并返回结果，即
    Figure PCTCN2022112499-appb-100003
    时，
    Figure PCTCN2022112499-appb-100004
    否则为0；
    Figure PCTCN2022112499-appb-100005
    为第l个时隙在本地处理的任务的集合，
    Figure PCTCN2022112499-appb-100006
    表示任务在本地处理；
    Figure PCTCN2022112499-appb-100007
    为第l个时隙用户任务卸载的集合，
    Figure PCTCN2022112499-appb-100008
    表示用户m的任务卸载至卫星n处理；
    Figure PCTCN2022112499-appb-100009
    为第l个时隙用户任务的调度集合，
    Figure PCTCN2022112499-appb-100010
    表示从卫星的传输或处理队列中调度用户m的任务，否则不调度任务；C l＝{C 1,l,C 2,l,...,C N,l},
    Figure PCTCN2022112499-appb-100011
    为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
    Figure PCTCN2022112499-appb-100012
    为第l个时隙卫星给各用户分配的计算资源；
    Figure PCTCN2022112499-appb-100013
    和
    Figure PCTCN2022112499-appb-100014
    分别为第l个时隙卫星n可分配的通信资源和计算资源；
    Figure PCTCN2022112499-appb-100015
    和
    Figure PCTCN2022112499-appb-100016
    分别为第l个时隙从卫星n的传输和处理队列中调度的任务集合；
    s.t.表示约束条件，约束条件式(18a)保证第l个时隙卫星给用户m分配的通信资源
    Figure PCTCN2022112499-appb-100017
    不应大于第l个时隙卫星n可分配的通信资源
    Figure PCTCN2022112499-appb-100018
    约束条件式(18b)保证第l个时隙卫星给用户m分配的计算资源
    Figure PCTCN2022112499-appb-100019
    不应大于第l个时隙卫星n可分配的计算资源
    Figure PCTCN2022112499-appb-100020
    约束条件式(18c)保证用户m的链路时延
    Figure PCTCN2022112499-appb-100021
    不应大于基于SNC设定的栅格i的限定时延
    Figure PCTCN2022112499-appb-100022
  3. 根据权利要求2所述的卫星传输优化方法，其特征在于，在第l个时隙，设卸载决策
    Figure PCTCN2022112499-appb-100023
    和调度决策
    Figure PCTCN2022112499-appb-100024
    已知，卫星的公平通信与计算资源分配子问题表述为：
    Figure PCTCN2022112499-appb-100025
    Figure PCTCN2022112499-appb-100026
    其中，
    Figure PCTCN2022112499-appb-100027
    为栅格i范围的用户m的任务从卫星n的传输队列中调度并分配通信资源后产生的传输时延；
    Figure PCTCN2022112499-appb-100028
    为栅格i范围的用户m的任务从卫星n的处理队列中调度并分配计算资源后产生的处理时延；
    Figure PCTCN2022112499-appb-100029
    为基于SNC设定的栅格i的限定时延；C l＝{C 1,l,C 2,l,...,C N,l},
    Figure PCTCN2022112499-appb-100030
    为第l个时隙卫星给各用户分配的通信资源；X l＝{X 1,l,X 2,l,...,X N,l}，
    Figure PCTCN2022112499-appb-100031
    为第l个时隙卫星给各用户分配的计算资源；
    Figure PCTCN2022112499-appb-100032
    和
    Figure PCTCN2022112499-appb-100033
    分别为第l个时隙卫星n可分配的通信资源和计算资源；
    s.t.表示约束条件，约束条件式(22a)保证第l个时隙卫星n给用户m分配的通信资源
    Figure PCTCN2022112499-appb-100034
    不应大于第l个时隙卫星n可分配的通信资源
    Figure PCTCN2022112499-appb-100035
    约束条件式(22b)保证用户m的链路时延
    Figure PCTCN2022112499-appb-100036
    不应大于基于SNC设定的栅格i的限定时延
    Figure PCTCN2022112499-appb-100037
    约束条件式(23a)保证第l个时隙卫星给用户m分配的计算资源
    Figure PCTCN2022112499-appb-100038
    不应大于第l个时隙卫星n可分配的计算资源
    Figure PCTCN2022112499-appb-100039
    约束条件式(23b)保证用户m的链路时延
    Figure PCTCN2022112499-appb-100040
    不应大于基于SNC设定的栅格i的限定时延
    Figure PCTCN2022112499-appb-100041
  4. 根据权利要求3所述的卫星传输优化方法，其特征在于，采用对偶上升法对公平通信与计算资源分配子问题进行求解，包括：
    采用对偶上升法对通信资源分配子问题进行求解：
    引入辅助变量χ，传输时延
    Figure PCTCN2022112499-appb-100042
    其中T m为用户m的任务k的大小；
    公式(22)转换为：
    Figure PCTCN2022112499-appb-100043
    Figure PCTCN2022112499-appb-100044
    构造拉格朗日函数A：
    Figure PCTCN2022112499-appb-100045
    其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子；
    则A的对偶函数为：
    Figure PCTCN2022112499-appb-100046
    其中，
    Figure PCTCN2022112499-appb-100047
    D的最大值即为公式(22)所求的最小值；
    通过交替迭代辅助变量χ和拉格朗日乘子μ m、ν、η，得到公平的通信资源分配
    Figure PCTCN2022112499-appb-100048
  5. 根据权利要求3所述的卫星传输优化方法，其特征在于，采用对偶上升法对公平通信与计算资源分配子问题进行求解，包括：
    采用对偶上升法对计算资源分配子问题进行求解：
    引入辅助变量χ，处理时延
    Figure PCTCN2022112499-appb-100049
    其中T m为用户m的任务k的大小；
    公式(23)转换为：
    Figure PCTCN2022112499-appb-100050
    Figure PCTCN2022112499-appb-100051
    构造拉格朗日函数A：
    Figure PCTCN2022112499-appb-100052
    其中，μ m≥0，ν≥0,η≥0为拉格朗日乘子；
    则A的对偶函数为：
    Figure PCTCN2022112499-appb-100053
    其中，
    Figure PCTCN2022112499-appb-100054
    D的最大值即为公式(23)所求的最小值；
    通过交替迭代辅助变量χ和拉格朗日乘子μ m、ν、η，得到公平的计算资源分配
    Figure PCTCN2022112499-appb-100055
  6. 根据权利要求1所述的卫星传输优化方法，其特征在于，采用深度Q网络对任务卸载与调度决策子问题进行求解，包括：
    将物联网终端的信息：P U(l)、 X U,l、T l，卫星的信息：P s(l)、 Β l、 C l、 X l，队列的信息：Q trans,l、Q pro,l输入预训练好的深度Q网络；
    得到输出的任务的卸载决策
    Figure PCTCN2022112499-appb-100056
    和调度决策
    Figure PCTCN2022112499-appb-100057
    定义第l个时隙的状态h l＝{P U(l),P s(l),T l, Β l, X U,l, C l, X l,Q trans,l,Q pro,lQ local,l}，P U(l)，P s(l)分别为第l个时隙用户和卫星的位置；T l＝{T 1,l,T 2,l,...,T M,l}为第l个时隙等待或正在调度的用户任务的大小； Β l＝{ Β 1,l, Β 2,l,..., Β M,l}为第l个时隙用户关联卫星的情况， Β m,l∈{1,2...,N}； X U,l＝{ X 1,l, X 2,l,..., X M,l}， X m,l∈{0,1}表示第l个时隙用户本地的计算资源是否被占用，即是否有任务正在本地处理； C l＝{ C 1,l, C 2,l,..., C N,l}为第l个时隙卫星被占用的通信资源； X l＝{ X 1,l, X 2,l,..., X N,l}为第l个时隙卫星被占用的计 算资源；
    Figure PCTCN2022112499-appb-100058
    和
    Figure PCTCN2022112499-appb-100059
    分别为第l个时隙卫星的传输队列和处理队列中等待调度的任务总量；
    Figure PCTCN2022112499-appb-100060
    为第l个时隙本地用户任务到达队列中等待调度的任务总量；
    从任务卸载和调度决策的角度定义第l个时隙的动作
    Figure PCTCN2022112499-appb-100061
    Figure PCTCN2022112499-appb-100062
    其中，E off∈{1,2...,Z}为任务的卸载决策，E off＝0表示任务在本地处理，E off＝Z表示任务卸载至卫星Z处理，E exe∈{0,1}表示是否从卫星的传输或处理队列中调度任务；通过动作a l即可得到用户当前任务的卸载决策
    Figure PCTCN2022112499-appb-100063
    和调度决策
    Figure PCTCN2022112499-appb-100064
  7. 根据权利要求1所述的卫星传输优化方法，其特征在于，栅格时延边界表示为：
    Figure PCTCN2022112499-appb-100065
    其中D(n)为时延函数，P(D(n)＞x)表示时延超过某个值x的概率，β(n)为随机服务曲线，β(n)是关于n的线性函数，即β(n)＝ρ β(θ)n，ρ β是随机服务曲线β(n)的斜率函数，θ、θ 1均为大于0的自由参数。
  8. 一种卫星传输优化装置，其特征在于，包括处理器及存储介质；
    所述存储介质用于存储指令；
    所述处理器用于根据所述指令进行操作以执行根据权利要求1至7任一项所述方法的步骤。
  9. 一种存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至7任一项所述方法的步骤。
PCT/CN2022/112499 2022-06-20 2022-08-15 一种基于栅格时延预测的卫星传输优化方法 WO2023245835A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210694561.0 2022-06-20
CN202210694561.0A CN115276754B (zh) 2022-06-20 2022-06-20 一种基于栅格时延预测的卫星传输优化方法

Publications (1)

Publication Number Publication Date
WO2023245835A1 true WO2023245835A1 (zh) 2023-12-28

Family

ID=83761527

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/112499 WO2023245835A1 (zh) 2022-06-20 2022-08-15 一种基于栅格时延预测的卫星传输优化方法

Country Status (2)

Country Link
CN (1) CN115276754B (zh)
WO (1) WO2023245835A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117595918A (zh) * 2024-01-18 2024-02-23 北京邮电大学 一种基于优先级的星间切换的方法及通信卫星
CN117829277A (zh) * 2024-03-04 2024-04-05 中国人民解放军国防科技大学 面向多类型卫星的应急观测任务调度方法、系统及装置
CN117915405A (zh) * 2024-03-18 2024-04-19 华东交通大学 一种分布式多无人机协同任务卸载方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6553225B1 (en) * 1999-02-03 2003-04-22 Wei Zhao Apparatus and method for positioning single-offset zones in a spot beam coverage area to maximize call duration in a satellite-based communications network
CN106571883A (zh) * 2016-07-04 2017-04-19 长春理工大学 卫星网络性能评价的随机网络演算方法
CN111148161A (zh) * 2020-01-06 2020-05-12 中国科学院微小卫星创新研究院 低轨卫星星座星间负载路由均衡方法及系统
CN113242568A (zh) * 2021-06-08 2021-08-10 重庆邮电大学 一种不确定网络环境中的任务卸载和资源分配方法
CN113377531A (zh) * 2021-06-04 2021-09-10 重庆邮电大学 基于无线能量驱动的移动边缘计算分布式服务部署方法
CN113434212A (zh) * 2021-06-24 2021-09-24 北京邮电大学 基于元强化学习的缓存辅助任务协作卸载与资源分配方法
CN114615705A (zh) * 2022-03-11 2022-06-10 广东技术师范大学 一种基于5g网络下单用户资源分配策略方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10069570B2 (en) * 2016-01-27 2018-09-04 Ciena Corporation Multi-layer modem reclamation systems and methods
CN113258985B (zh) * 2021-05-08 2023-01-31 中国电子科技集团公司第五十四研究所 单站多星mimo上注系统能效优化方法
CN113258988B (zh) * 2021-05-13 2022-05-20 重庆邮电大学 一种基于dqn的多业务低轨卫星资源分配方法
CN114362810B (zh) * 2022-01-11 2023-07-21 重庆邮电大学 一种基于迁移深度强化学习的低轨卫星跳波束优化方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6553225B1 (en) * 1999-02-03 2003-04-22 Wei Zhao Apparatus and method for positioning single-offset zones in a spot beam coverage area to maximize call duration in a satellite-based communications network
CN106571883A (zh) * 2016-07-04 2017-04-19 长春理工大学 卫星网络性能评价的随机网络演算方法
CN111148161A (zh) * 2020-01-06 2020-05-12 中国科学院微小卫星创新研究院 低轨卫星星座星间负载路由均衡方法及系统
CN113377531A (zh) * 2021-06-04 2021-09-10 重庆邮电大学 基于无线能量驱动的移动边缘计算分布式服务部署方法
CN113242568A (zh) * 2021-06-08 2021-08-10 重庆邮电大学 一种不确定网络环境中的任务卸载和资源分配方法
CN113434212A (zh) * 2021-06-24 2021-09-24 北京邮电大学 基于元强化学习的缓存辅助任务协作卸载与资源分配方法
CN114615705A (zh) * 2022-03-11 2022-06-10 广东技术师范大学 一种基于5g网络下单用户资源分配策略方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117595918A (zh) * 2024-01-18 2024-02-23 北京邮电大学 一种基于优先级的星间切换的方法及通信卫星
CN117595918B (zh) * 2024-01-18 2024-05-07 北京邮电大学 一种基于优先级的星间切换的方法及通信卫星
CN117829277A (zh) * 2024-03-04 2024-04-05 中国人民解放军国防科技大学 面向多类型卫星的应急观测任务调度方法、系统及装置
CN117915405A (zh) * 2024-03-18 2024-04-19 华东交通大学 一种分布式多无人机协同任务卸载方法

Also Published As

Publication number Publication date
CN115276754A (zh) 2022-11-01
CN115276754B (zh) 2023-06-16

Similar Documents

Publication Publication Date Title
WO2023245835A1 (zh) 一种基于栅格时延预测的卫星传输优化方法
Seid et al. Collaborative computation offloading and resource allocation in multi-UAV-assisted IoT networks: A deep reinforcement learning approach
Liu et al. RL/DRL meets vehicular task offloading using edge and vehicular cloudlet: A survey
Kopeikin et al. Dynamic mission planning for communication control in multiple unmanned aircraft teams
Tang et al. Federated learning for intelligent transmission with space-air-ground integrated network (SAGIN) toward 6G
EP4024212B1 (en) Method for scheduling inference workloads on edge network resources
Wang et al. Fuzzy-CNN based multi-task routing for integrated satellite-terrestrial networks
Yang et al. Multi-armed bandits learning for task offloading in maritime edge intelligence networks
Guo et al. Predictive resource allocation with deep learning
Han et al. Time-varying topology model for dynamic routing in LEO satellite constellation networks
Qin et al. Service-aware resource orchestration in ultra-dense LEO satellite-terrestrial integrated 6G: A service function chain approach
Zhu et al. Load-balanced virtual network embedding based on deep reinforcement learning for 6G regional satellite networks
Nguyen et al. Deep reinforcement learning-based partial task offloading in high altitude platform-aided vehicular networks
Fu et al. Toward energy-efficient UAV-assisted wireless networks using an artificial intelligence approach
Cai et al. Mobile edge computing network control: Tradeoff between delay and cost
Bozkaya et al. Airnet: Energy-aware deployment and scheduling of aerial networks
Zhou et al. Joint multi-objective optimization for radio access network slicing using multi-agent deep reinforcement learning
Liu et al. Learning based fluctuation-aware computation offloading for vehicular edge computing system
Mostafa et al. Machine Learning-Based Multi-UAVs Deployment for Uplink Traffic Sizing and Offloading in Cellular Networks
Fraire et al. On the design of fair contact plans in predictable delay-tolerant networks
Zhang et al. Two-timescale online learning of joint user association and resource scheduling in dynamic mobile edge computing
Sobouti et al. Managing sets of flying base stations using energy efficient 3D trajectory planning in cellular networks
CN116828548A (zh) 一种针对电力无线网络的基于强化学习的最佳路由调度方法
Meng et al. Intelligent routing orchestration for ultra-low latency transport networks
Matskani et al. The mutual benefits of primary-secondary user cooperation in wireless cognitive networks

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22947576

Country of ref document: EP

Kind code of ref document: A1