CN115659803A - Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition - Google Patents
Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition Download PDFInfo
- Publication number
- CN115659803A CN115659803A CN202211311838.3A CN202211311838A CN115659803A CN 115659803 A CN115659803 A CN 115659803A CN 202211311838 A CN202211311838 A CN 202211311838A CN 115659803 A CN115659803 A CN 115659803A
- Authority
- CN
- China
- Prior art keywords
- unmanned aerial
- aerial vehicle
- intelligent terminal
- twin
- unloading
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention discloses a method, a system and a computer readable storage medium for intelligently unloading a computing task under the condition of twin network mapping error of an unmanned aerial vehicle, wherein the method comprises the following steps: s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data; s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target; s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy; s4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3). The invention reduces unreliable remote communication between the terminal user and the edge server, reduces system delay and enhances the practicability of data.
Description
Technical Field
The invention relates to the technical field of unmanned aerial vehicles, in particular to a method and a system for intelligently unloading a computing task under the condition of twin network mapping error of an unmanned aerial vehicle and a computer-readable storage medium.
Background
The internet of things equipment and the application thereof are increasingly popularized, more and more applications requiring resources and novel big data services are developing and playing a role in our daily life. However, terminal devices have difficulties in serving delay sensitive and compute intensive applications due to limited physical size, battery capacity, and computational resources. Mobile Edge Computing (MEC) has recently been considered a promising technology to serve resource-limited terminals at the edge of a wireless network. And the terminal unloads the calculation intensive tasks to the MEC server for execution through the calculation unloading service. Therefore, most of the energy consumption of task execution is shifted from resource-limited terminals to resource-rich MEC servers. Thereby reducing the computing delay and energy consumption of tasks, prolonging the service life of the whole network and improving the service quality of the network. However, when a large number of tasks are offloaded to the MEC server at the same time, severe resource contention may occur on the MEC server, resulting in network congestion and performance degradation. And the end-to-end (D2D) cooperation is considered as a solution to the potential problem of resource shortage, and idle computing resources are provided by idle mobile users and terminal devices to perform computation of the offloading task, so that delay can be effectively reduced. However, the MEC technique does not support flexible deployment, and in some environments with sparsely distributed communication infrastructure and poor communication conditions, such as plateau, forest and other areas, it is still difficult for the terminal to obtain reliable computing services.
With the continuous development and maturity of the unmanned aerial vehicle technology and the advantages of flexible deployment and high working efficiency of the unmanned aerial vehicle, the industrial-grade unmanned aerial vehicle is widely applied, such as application in disaster relief, electric power, aerial survey and the like. Meanwhile, the unmanned aerial vehicle has been proposed as a 5G and after-5G in-flight MEC server to extend wireless coverage and serve terminals in uncertain and extreme environments, and provide computation offload service with low network overhead and execution delay. To perform efficient and reliable computation offload services and provide ubiquitous communication and computation support, the network edge computation of the unmanned aerial vehicle requires joint optimization of computation task offload, local computation, unmanned aerial vehicle deployment, and unmanned aerial vehicle flight trajectory. The traditional optimization method depends on human experience to carry out optimization decision, and the labor cost is high. With the recent massive application of artificial intelligence technology in various industries, machine learning and artificial intelligence algorithms applied in the field of mobile Network optimization are also increasing, for example, a multi-objective optimization model is used to optimize Network capacity and coverage rate, a Neural Network is used to predict the target position of a user in mobility management, and near real-time resource allocation is realized by a Deep Neural Network (DNN). However, these methods cannot adapt to dynamic changes in the state of the high dimensional network and the results are often sub-optimal. Therefore, some studies have adopted Reinforcement Learning (RL) to solve this problem. And the reinforcement learning is based on the feedback reward after the system state transition and action, and the optimal decision under the unknown environment is learned. It can interact with complex and dynamic environments to optimize learning objectives that are difficult to model. However, the difficulty of obtaining training samples and the time cost are significant challenges. To do this, these algorithms interact directly with the physical network in the exploration process, which means that they apply the actions resulting from the current untrained strategy directly into the network to obtain training samples. This is particularly dangerous because current suboptimal or even poor actions may deteriorate network performance, which may cause huge losses in the drone network that are difficult to recover.
The prior art discloses an unmanned aerial vehicle cluster trajectory optimization and task unloading method based on digital twins, which comprises the following steps: constructing an unmanned aerial vehicle cluster auxiliary edge calculation model; constructing a physical entity network; constructing a digital twin network of a physical entity network, and fitting the geographic positions and the resource state information of the user equipment and the unmanned aerial vehicle; constructing an optimization model of unmanned aerial vehicle track, user equipment unloading decision and computing resource allocation; solving the unmanned aerial vehicle track, the user equipment and the calculation resource allocation strategy of the unmanned aerial vehicle; obtaining an offloading decision of a user equipment; and obtaining the track of the unmanned aerial vehicle, the optimal allocation strategy of the user equipment and the computing resources of the unmanned aerial vehicle, and obtaining the optimal unloading decision of the computing task of the user equipment. In the scheme, two algorithms are adopted for optimization, so that the complexity is high and the stability is poor.
In order to solve the problems, the invention provides a method and a system for intelligently unloading a computing task under the condition of twin network mapping errors of an unmanned aerial vehicle, and a computer-readable storage medium.
Disclosure of Invention
The invention provides a method and a system for intelligently unloading a calculation task under the condition of twin network mapping error of an unmanned aerial vehicle and a computer readable storage medium, which can reduce unreliable remote communication between an end user and an edge server, reduce system delay and enhance the practicability of data.
The primary objective of the present invention is to solve the above technical problems, and the technical solution of the present invention is as follows:
the invention provides a method for intelligently unloading a computing task under the condition of twin network mapping error of an unmanned aerial vehicle, which comprises the following steps: an intelligent unloading method for a computing task under the condition of twin network mapping error of an unmanned aerial vehicle comprises the following steps:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
s4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3).
Further, the digital twin model expression of the intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)} (1)
wherein, M i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the updating state of the intelligent terminal.
Further, the twin model expression of the unmanned aerial vehicle is as follows:
DT j ＝{M j ,D j ,s j (t),s j (t+1)} (2)
wherein, M j Behavioral model representing unmanned aerial vehicle, s j (t) represents the real-time status of the drone, s j (t + 1) represents the update status of the drone.
Further, constructing a calculation unloading and resource allocation strategy, and taking the maximum utility function of the unmanned aerial vehicle as an optimization target comprises: defining a scene, a communication model and a calculation unloading model;
defining a scene:
setting I intelligent terminals which are randomly distributed and recorded as U = {1,2, \8230 =, I }, wherein each intelligent terminal can execute a calculation task locally and unload the calculation task to an unmanned aerial vehicle carrying an edge server for execution, setting that the unmanned aerial vehicle flies to a specified suspension point from the current position to execute the calculation task, and only one intelligent terminal is served by the unmanned aerial vehicle at the same time, setting the maximum service duration of the unmanned aerial vehicle as T, wherein the unmanned aerial vehicle is positioned on a horizontal plane with the height of H in the process of providing the calculation service, and at the T moment, T belongs to {0,1,2,...,. T }, and the decision variable of the k-th suspension point is selected as a by the unmanned aerial vehicle k Wherein K is equal to {0,1, 2.., K }, when a k When =1, the unmanned aerial vehicle is indicated to select the kth suspension point, and at this time, the position of the unmanned aerial vehicle is indicated as
On the contrary, a k =0 indicates that unmanned aerial vehicle stays at present position, w for ground intelligent terminal i's position i ＝(x i ,y i ) Representing;
defining a communication model:
at the t moment, the unmanned aerial vehicle hovers at the kth suspension point, the channel bandwidth is defined as B, the ground intelligent equipment alternately unloads the calculation task to the unmanned aerial vehicle, and the communication rate of the ith intelligent equipment and the unmanned aerial vehicle at the kth suspension point is represented as:
wherein p is i For the transmission power of the ith smart device, σ represents the noise power,
representing the channel gain between the ith intelligent terminal and the unmanned aerial vehicle, wherein beta is fixed transmission loss, m represents a path loss factor, and d i,k Representing the communication distance from the k-th suspension point of the unmanned aerial vehicle to the user i;
defining a computational offload model:
defining a local computation delay of the ith intelligent device as
If the ith intelligent device unloads the calculation task to the unmanned aerial vehicle, defining the calculation unloading time delay as follows:
wherein
Which represents the time delay of the data transmission,
representing the calculated time delay, f i,k The computational resources allocated to the intelligent device i for the drone, definition a i,k For offloading decisions, when a i.k If =1, it means that the ith intelligent device unloads the computation task to the unmanned aerial vehicle hovering at the kth suspension point, otherwise a i.k When =0, it indicates that the ith intelligent device will perform task calculation locally, and since the unmanned aerial vehicle performs the task calculation for unloading in a hovering manner, the hovering time of the unmanned aerial vehicle is equal to the calculated unloading time delay, that is, the hovering time of the unmanned aerial vehicle at the suspension point k is represented as:
the drone needs to be propelled from the hover point k' to the current hover point k position in the horizontal direction, so the flying energy consumption of the drone is expressed as:
wherein P is 1 For propulsion power, V is the flight speed of the drone, and according to equation (5), the hovering energy consumption of the drone at the kth suspension point is expressed as:
E h (t)＝P 2 T k (7)
in the formula P 2 Hovering power for unmanned aerial vehicle, T k For hover time, the energy consumption of the drone to perform the offload task computation is expressed as:
in summary, according to equations (6), (7) and (8), the total energy consumption of the drone during the calculation of the unloading process is represented as:
calculating and processing task R of unmanned aerial vehicle i The utility function of (2) is defined as:
wherein a is i Indicates that the unmanned aerial vehicle has the calculation processing size of H i (t) price per unit of task, b i Representing the price of the unmanned aerial vehicle unit computing resource, and u representing the price of the unmanned aerial vehicle unit cost, then, taking the maximized unmanned aerial vehicle utility function as an optimization target to be represented as follows:
C 2 ：E f (t)+E h (t)+E c (t)<RE(t-1),
C 3 :a k ∈(0,1),a i,k ∈(0,1),
C 5 :0≤f i,k ≤f max ,
wherein C1 represents that the total energy consumption of the unmanned aerial vehicle in the calculation unloading process can not exceed the maximum energy E of the battery of the unmanned aerial vehicle d (ii) a C2 represents that the calculated unloading total energy consumption of the unmanned aerial vehicle at the current moment is less than the residual energy RE (t-1) of the unmanned aerial vehicle battery at the last moment; c3 represents a hovering decision of the unmanned aerial vehicle and an unloading decision of the ground intelligent terminal i; c4 represents that the ground intelligent terminal i can only unload the task to one unmanned aerial vehicle; each unmanned aerial vehicle can only provide calculation unloading service for one ground intelligent terminal at the same time; c5 represents that the computing resource allocated to the ground intelligent terminal i by the unmanned aerial vehicle cannot exceed the maximum computing resource f of the unmanned aerial vehicle max (ii) a C6 represents that the task time delay calculated by the intelligent terminal i does not exceed the maximum tolerance time delay.
Further, constructing the optimization objective as a markov decision process using a competitive near-end strategy optimization method comprises: defining a state space, an action space and a reward function,
state space: at decision time T, T is from {0,1, \8230;, T }, the twin state space of the local base station is defined as:
wherein
Representing drone position, RE (t) representing drone residual energy, w i (t) represents the position of the ground intelligent terminal i, R i (t) represents the calculation task of the ground intelligent terminal i, h i,k (t) represents the channel gain between the ground intelligent terminal I and the UAV hovering at the kth suspension point, wherein I belongs to {1,2, \8230;, I }, and K belongs to {0,1,.., K }
An action space: at decision time t, the action space that optimizes the utility of the drone is represented as:
t＝{a k ,a i,k ,f i,k } (13)
wherein a is k Represents the decision of the drone to select the kth suspension point at the time t, a i,k Representing the offloading decision of the ground intelligent terminal i, f i,k The unmanned aerial vehicle distributes computing resources of a ground intelligent terminal i;
the reward function: in the process of executing the near-end strategy optimization algorithm, whether the constraints C1-C6 are established or not needs to be checked, so that the following instant rewarding function r is designed t ：
If the constraints C1-C6 cannot be met, indicating that the performance of the current optimization decision is poor, in this case, setting the instant reward to 0 to avoid invalid decisions; in the algorithm training process, the local base station BS obtains an initial state from a maintained twin model, generates a random strategy, obtains a new state and an action transfer sequence through interaction with the twin model, and simultaneously adjusts respective Actor-critical network parameters to obtain an optimal strategy. The loss function of the Actor network parameter update procedure is expressed as:
wherein the ratio of probabilities
Indicating the magnitude of the update of the network parameter,
the value of the action is reflected by the dominance function, the epsilon is the clipping rate, and the aim of the clip function is to compare the probability ratio r t (theta) is limited to the interval [ 1-epsilon, 1 +. Epsilon]In the formula (1-3), when the merit function is
When r is t Stopping updating when (theta) exceeds 1 +. Epsilon
When r is t (theta) less than 1-epsilon stops the update in order to prevent the updated network from being too different from the old network, thus achieving a small stride update.
Further, the definition of the merit function is as follows:
δ t ＝r t +V(s t+1 )-V(s t ) (16)
wherein, the first and the second end of the pipe are connected with each other,
representing the merit function, δ t Represents a single step time error, gamma ∈ (0)1) denotes the discount coefficient, V(s) t+1 ) Represents a state s t+1 Estimated value of, V(s) t ) Represents a state s t The estimated value of (c).
Further, the criticic network calculates the state value by using the action value output by the sub-network, and the specific expression is as follows:
V(s t )＝Q(s t ,a) T P(s t ,a) (17)
wherein, P(s) t ,a)＝[p 1 ,p 1 ,…p K ]Respectively take action p 1 ,p 1 ,…p K Probability of (a), P(s) t A) is derived from the operator network output, while the Critic network will be delta t The mean square error of the network parameter is used as a loss function, the loss function is minimized by using a gradient descent method, and the network parameter is updated.
The invention provides a system for intelligently unloading a computing task under the condition of twin network mapping error of an unmanned aerial vehicle, which comprises: the intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition comprises a storage and a processor, wherein the storage comprises a program of the intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition, and when the program of the intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition is executed by the processor, the following steps are realized:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
s4: and (4) making a flight track, calculating unloading and distributing resources thereof by using the decision action output in the step (S3).
Further, the digital twin model expression of the intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)}
wherein M is i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the update state of the intelligent terminal.
The third aspect of the invention provides a computer-readable storage medium, wherein the computer-readable storage medium comprises a program of a method for intelligently unloading a calculation task under the condition of twin network mapping error of an unmanned aerial vehicle, and when the program of the method for intelligently unloading the calculation task under the condition of twin network mapping error of the unmanned aerial vehicle is executed by a processor, the method for intelligently unloading the calculation task under the condition of twin network mapping error of the unmanned aerial vehicle is realized.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
according to the invention, through constructing the digital twin models of the intelligent terminal and the unmanned aerial vehicle, the unmanned aerial vehicle network is displayed in a local BS in a digital twin form, the simulation and simulation generated twin data is combined with a competition near-end strategy to obtain a competition network, less energy consumption and calculation resources are obtained to obtain an optimal task unloading and resource allocation strategy, and the overall utility of the network is improved.
Drawings
Fig. 1 is a diagram of an unmanned aerial vehicle edge unloading scene based on digital twinning according to an embodiment of the present invention.
FIG. 2 is a flowchart of a method for intelligently unloading a computing task under the condition of a twin network mapping error of an unmanned aerial vehicle.
FIG. 3 is a block diagram of an intelligent unloading system for a computing task under the condition of a twin network mapping error of an unmanned aerial vehicle.
Detailed Description
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention, taken in conjunction with the accompanying drawings and detailed description, is set forth below. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described herein, and therefore the scope of the present invention is not limited by the specific embodiments disclosed below.
Example 1
As shown in fig. 1, which is a scenario of an embodiment of the present invention, a Base Station (BS) covers a certain range, and a certain number of users exist in the coverage area, where the users may be mobile devices, internet of things devices, and the like. Due to the problems of insufficient computing resources, insufficient energy and the like of personal users and Internet of things equipment, the users can choose to locally compute own computing tasks or request the unmanned aerial vehicles to compute and unload the tasks. Because the base station overload may be caused by directly transmitting data to the base station, and the unmanned aerial vehicle carries out auxiliary task unloading, the method and the device use a competition-near-end strategy optimization algorithm for training to obtain an optimal decision. Because the unloading decision needs to be tested continuously, and the test in the physical environment has risks, the invention combines the digital twin technology, each user and the unmanned aerial vehicle generate own digital twin bodies, the base station is responsible for maintaining the digital twin bodies and forming a digital twin network, and the near-end strategy optimization algorithm is trained and tested in the digital twin network until acting on the real unmanned aerial vehicle network after convergence. According to the process, the method can make an optimal decision in the twin network, so that the time delay of the calculation task of the unmanned aerial vehicle network is reduced, the energy consumption of the system is reduced, and the overall utility of the network is improved.
As shown in fig. 2, a first aspect of the present invention provides a method for intelligently unloading a computation task under a twin network mapping error of an unmanned aerial vehicle, including the following steps:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
it should be noted that, in the present invention, the wireless digital twin network system is divided into a wireless access layer (i.e., a terminal layer) and a digital twin layer (i.e., an edge layer). The wireless access layer is composed of entities such as terminal equipment with limited computing and storage resources, unmanned aerial vehicles and the like. Through wireless communication, these entities connect to nearby base stations to construct a digital twin. In the digital twin layer, a base station is provided with an MEC server to continuously acquire data of the intelligent terminal and the unmanned aerial vehicle in the service area, and calculates and analyzes behavior characteristics of the intelligent terminal and the unmanned aerial vehicle so as to construct a twin model of the intelligent terminal, the unmanned aerial vehicle, a network topological relation and the environment in the service area. The digital twin of the internet of things device is a complete copy of the physical device, including hardware configuration information, historical operating data, and real-time status.
The expression of the digital twin model of the intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)} (1)
wherein M is i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the update state of the intelligent terminal.
The twin model expression of the unmanned aerial vehicle is as follows:
DT j ＝{M j ,D j ,s j (t),s j (t+1)} (2)
wherein M is j Representing a behavioral model of the drone, s j (t) represents the real-time status of the drone, s j (t + 1) represents the update status of the drone.
And simulating the running states of the intelligent terminal, the unmanned aerial vehicle and the like in the service area through the twin model and generating twin data.
S2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
the method specifically comprises the following steps: defining a scene, a communication model and a calculation unloading model;
defining a scene:
setting I intelligent terminals which are randomly distributed and marked as U = {1,2, \ 8230:, I }, wherein each intelligent terminal can execute a calculation task locally and also can unload the calculation task to an unmanned aerial vehicle carrying an edge server for execution, setting that the unmanned aerial vehicle flies from the current position to a specified suspension point to execute the calculation task, and only one intelligent terminal is served by the unmanned aerial vehicle at the same time, setting the maximum service duration of the unmanned aerial vehicle as T, wherein the unmanned aerial vehicle is positioned on a horizontal plane with the height of H in the process of providing the calculation service, and at the T moment, T belongs to {0,1,2,.. And T }, and the decision variable of the k-th suspension point is selected as a by the unmanned aerial vehicle k Wherein K is equal to {0,1, 2.., K }, when a k When =1, the unmanned aerial vehicle is indicated to select the kth suspension point, and at this time, the position of the unmanned aerial vehicle is indicated as
On the contrary, a k =0 indicates that unmanned aerial vehicle stays at present position, w for position of ground intelligent terminal i i ＝(x i ,y i ) Represents;
defining a communication model:
at the t moment, the unmanned aerial vehicle hovers at the kth suspension point, the channel bandwidth is defined as B, the ground intelligent equipment alternately unloads the calculation task to the unmanned aerial vehicle, and the communication rate of the ith intelligent equipment and the unmanned aerial vehicle at the kth suspension point is represented as:
wherein p is i For the transmission power of the ith smart device, σ represents the noise power,
represents the channel gain between the ith intelligent terminal and the unmanned aerial vehicle, wherein beta is the fixed transmission loss, m represents the path loss factor, d i,k Representing the communication distance from the k-th suspension point of the unmanned aerial vehicle to the user i;
defining a computational offload model:
defining a local computation delay of the ith intelligent device as
If the ith intelligent device unloads the calculation task to the unmanned aerial vehicle, defining the calculation unloading time delay as follows:
wherein
Which is indicative of the time delay of the data transmission,
representing the calculated time delay, f i,k The computational resources allocated to the intelligent device i for the drone, definition a i,k For offloading decisions, when a i.k If =1, it means that the ith intelligent device unloads the computation task to the unmanned aerial vehicle hovering at the kth suspension point, otherwise a i.k When =0, it indicates that the ith intelligent device will perform task calculation locally, and since the unmanned aerial vehicle performs the task calculation for unloading in a hovering manner, the hovering time of the unmanned aerial vehicle is equal to the calculated unloading time delay, that is, the hovering time of the unmanned aerial vehicle at the suspension point k is represented as:
the drone needs to be propelled in the horizontal direction from the hover point k' to the position of the current hover point k, so the flight energy consumption of the drone is expressed as:
wherein P is 1 For propulsion power, V is the flight speed of the drone, according to equation (5) with the drone on the secondThe hover energy consumption for k hover points is expressed as:
E h (t)＝P 2 T k (7)
in the formula P 2 Hovering power for unmanned aerial vehicle, T k For hover time, the energy consumption of the drone to perform the offload task computation is expressed as:
in summary, according to equations (6), (7) and (8), the total energy consumption of the drone during the calculation of the unloading process is represented as:
unmanned aerial vehicle calculation processing task R i Is defined as:
wherein a is i Indicates that the unmanned aerial vehicle has the calculation processing size of H i (t) price per unit of task, b i Representing the price of the unmanned aerial vehicle unit computing resource, and u representing the price of the unmanned aerial vehicle unit cost, then taking the maximized unmanned aerial vehicle utility function as an optimization target to be represented as follows:
C 2 ：E f (t)+E h (t)+E c (t)<RE(t-1),
C 3 :a k ∈(0,1),a i,k ∈(0,1),
C 5 :0≤f i,k ≤f max ,
wherein C1 represents that the total energy consumption of the unmanned aerial vehicle in the calculation unloading process can not exceed the maximum energy E of the battery of the unmanned aerial vehicle d (ii) a C2 represents that the calculated unloading total energy consumption of the unmanned aerial vehicle at the current moment is less than the residual energy RE (t-1) of the unmanned aerial vehicle battery at the last moment; c3, representing a hovering decision of the unmanned aerial vehicle and an unloading decision of the ground intelligent terminal i; c4 represents that the ground intelligent terminal i can only unload the task to one unmanned aerial vehicle; each unmanned aerial vehicle can only provide calculation unloading service for one ground intelligent terminal at the same time; c5 represents that the computing resource distributed to the ground intelligent terminal i by the unmanned aerial vehicle cannot exceed the maximum computing resource f of the unmanned aerial vehicle max (ii) a C6 represents that the task delay calculated by the intelligent terminal i does not exceed the maximum tolerance delay.
S3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
in order to obtain the optimal solution of the optimization problem, the invention combines a competition network and a near-end strategy optimization algorithm and provides a competition-near-end strategy optimization algorithm based on a twin model. The optimization problem (11) is structured as a markov decision process, wherein the state space, the action space and the reward function are defined as follows:
state space: at decision time T, T is from {0,1, \8230;, T }, the twin state space of the local base station is defined as:
wherein
Representing drone position, RE (t) representing drone residual energy, w i (t) represents the position of the ground intelligent terminal i, R i (t) represents the calculation task of the ground intelligent terminal i, h i,k (t) represents the channel gain between the ground intelligent terminal I and the UAV hovering at the kth suspension point, wherein I belongs to {1,2, \8230;, I }, and K belongs to {0,1,.., K }
An action space: at decision time t, the action space that optimizes the utility of the drone is represented as:
t＝{a k ,a i,k ,f i,k } (13)
wherein a is k Represents the decision of the drone to select the kth hover point at the t-th instant, a i,k Representing the offloading decision of the ground intelligent terminal i, f i,k The unmanned aerial vehicle allocates the computing resources of the ground intelligent terminal i;
the reward function: in the process of executing the near-end strategy optimization algorithm, whether the constraints C1-C6 are established or not needs to be checked, so that the following instant rewarding function r is designed t ：
If the constraints C1-C6 cannot be met, indicating that the performance of the current optimization decision is poor, in this case, setting the instant reward to 0 to avoid invalid decisions; in the algorithm training process, the local base station BS obtains an initial state from a maintained twin model, generates a random strategy, obtains a new state and an action transfer sequence through interaction with the twin model, and simultaneously adjusts respective Actor-critical network parameters to obtain an optimal strategy. The loss function of the Actor network parameter update procedure is represented as:
wherein the ratio of probabilities
Indicating the magnitude of the update of the network parameter,
the value of an action is reflected by the representation advantage function, the epsilon is the clipping rate, and the aim of the clip function is to compare the probability ratio r t (theta) is limited to the interval [ 1-epsilon, 1 +. Epsilon]In the formula (1-3), when the merit function is
When r is t Stopping updating when (theta) exceeds 1+ epsilon
When r is t (theta) less than 1-epsilon stops the update in order to prevent the updated network from being too different from the old network, thus achieving a small stride update.
Further, the definition of the merit function is as follows:
δ t ＝r t +V(s t+1 )-V(s t ) (16)
wherein, among others,
representing the merit function, δ t Represents a single step time error, γ ∈ (0, 1) represents a discount coefficient, V(s) t+1 ) Represents a state s t+1 Estimated value of, V(s) t ) Represents a state s t The estimated value of (c).
The competitive network estimates an action value function more accurately by adding two sub-networks behind the DQN network, takes the state as the input of the competitive network, the neural network N1 extracts characteristic information, then the output is respectively taken as the input of the networks N2 and N3, and respectively outputs the state value V(s) and the dominant value A (s, a) of each action, the action value Q (s, a) of each action can be obtained by adding the state value V(s) and the dominant value A (s, a), the Q network obtains more efficient updating by splitting the observation of the action value into the observation of the state value and the observation of the action dominant value, and the action dominant value A (s, a) is required to be subtracted from the average value of the action dominant value in order to prevent the action dominant value A (s, a) from compensating the change of the action value Q (s, a) in the updating process.
S4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3).
It should be noted that, in the present invention, the competitive near-end policy optimization method is based on the Actor-Critic framework, and the update target of the Actor network is the same as the near-end policy optimization algorithm, so as to optimize the network parameters according to the minimization loss function in the formula (15). The Critic network adopts the competitive network structure shown in fig. 1, and since the accuracy of the dominance function value and the state value output by the sub-network is relatively low, the state value needs to be calculated by using the motion value of the output, as shown in the following formula:
V(s t )＝Q(s t ,a) T P(s t ,a) (17)
wherein P(s) t ,a)＝[p 1 ,p 1 ,…p K ]Respectively taking action p 1 ,p 1 ,…p K The probability of (c). P(s) t And a) is available from the operator network output. At the same time, the Critic network will be δ t The mean square error of the network parameter is used as a loss function, the loss function is minimized by using a gradient descent method, and the network parameter is updated. And continuously repeating the iterative training process until an optimal strategy is learned, making a flight trajectory according to the final decision action, calculating unloading and checking the distribution of resources in the twin network, and acting the decision on the physical entity after the flight trajectory passes the check.
The invention provides a system for intelligently unloading a computing task under the condition of twin network mapping error of an unmanned aerial vehicle, which comprises: the intelligent unloading method program for the calculation task under the unmanned aerial vehicle twin network mapping error condition is executed by the processor to realize the following steps:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
s4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3).
Further, the expression of the digital twin model of the intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)}
wherein, M i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the updating state of the intelligent terminal.
The third aspect of the invention provides a computer-readable storage medium, wherein the computer-readable storage medium comprises a program of a method for intelligently unloading a calculation task under the condition of twin network mapping error of an unmanned aerial vehicle, and when the program of the method for intelligently unloading the calculation task under the condition of twin network mapping error of the unmanned aerial vehicle is executed by a processor, the method for intelligently unloading the calculation task under the condition of twin network mapping error of the unmanned aerial vehicle is realized.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A method for intelligently unloading a calculation task under the condition of twin network mapping error of an unmanned aerial vehicle is characterized by comprising the following steps:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
s4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3).
2. The intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition as claimed in claim 1, wherein the digital twin model expression of the intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)} (1)
wherein, M i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the updating state of the intelligent terminal.
3. The intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition as claimed in claim 1, wherein the expression of the unmanned aerial vehicle twin model is as follows:
DT j ＝{M j ,D j ,s j (t),s j (t+1)} (2)
wherein M is j Behavioral model representing unmanned aerial vehicle, s j (t) represents the real-time status of the drone, s j (t + 1) represents the update status of the drone.
4. The method for intelligently offloading computing tasks under the situation of twin network mapping errors of unmanned aerial vehicles according to claim 1, wherein a computing offloading and resource allocation strategy is constructed, and maximizing an unmanned aerial vehicle utility function as an optimization objective comprises: defining a scene, a communication model and a calculation unloading model;
defining a scene:
setting I intelligent terminals which are randomly distributed and marked as U = {1,2, \ 8230:, I }, wherein each intelligent terminal can execute a calculation task locally and also can unload the calculation task to an unmanned aerial vehicle carrying an edge server for execution, setting that the unmanned aerial vehicle flies from the current position to a specified suspension point to execute the calculation task, and only one intelligent terminal is served by the unmanned aerial vehicle at the same time, setting the maximum service duration of the unmanned aerial vehicle as T, wherein the unmanned aerial vehicle is positioned on a horizontal plane with the height of H in the process of providing the calculation service, and at the T moment, T belongs to {0,1,2,.. And T }, and the decision variable of the k-th suspension point is selected as a by the unmanned aerial vehicle k Wherein K is equal to {0,1, 2.., K }, when a is k =1, the drone selects the kth suspension point, and the position of the drone is represented as
On the contrary, a k =0 indicates that unmanned aerial vehicle stays at present position, w for ground intelligent terminal i's position i ＝(x i ,y i ) Representing;
defining a communication model:
at the t moment, the unmanned aerial vehicle hovers at the kth suspension point, the channel bandwidth is defined as B, the ground intelligent equipment unloads calculation tasks to the unmanned aerial vehicle in turn, and the communication rate between the ith intelligent equipment and the unmanned aerial vehicle at the kth suspension point is represented as:
wherein p is i Is the transmission power of the ith intelligent device, sigmaWhich is indicative of the power of the noise,
representing the channel gain between the ith intelligent terminal and the unmanned aerial vehicle, wherein beta is fixed transmission loss, m represents a path loss factor, and d i,k Representing the communication distance from the k-th suspension point of the unmanned aerial vehicle to the user i;
defining a computational offload model:
defining a local computation delay of the ith intelligent device as
If the ith intelligent device unloads the calculation task to the unmanned aerial vehicle, defining the calculation unloading time delay as follows:
wherein
Which represents the time delay of the data transmission,
representing the calculated time delay, f i,k The computational resources allocated to the intelligent device i for the drone, definition a i,k For offloading decisions, when a i.k When =1, it indicates that the ith intelligent device offloads the computing task to the unmanned aerial vehicle hovering at the kth suspension point, otherwise, a i.k When =0, it indicates that the ith intelligent device will perform task calculation locally, and since the unmanned aerial vehicle performs the task calculation for unloading in a hovering manner, the hovering time of the unmanned aerial vehicle is equal to the calculated unloading time delay, that is, the hovering time of the unmanned aerial vehicle at the suspension point k is represented as:
the drone needs to be propelled from the hover point k' to the current hover point k position in the horizontal direction, so the flying energy consumption of the drone is expressed as:
wherein P is 1 For propulsion power, V is the flight speed of the drone, and according to equation (5), the hovering energy consumption of the drone at the kth suspension point is expressed as:
E h (t)＝P 2 T k (7)
in the formula P 2 Hovering power for unmanned aerial vehicle, T k For hover time, the energy consumption of the drone to perform the offload task computation is expressed as:
in summary, according to equations (6), (7) and (8), the total energy consumption of the drone during the calculation of the offloading is expressed as:
calculating and processing task R of unmanned aerial vehicle i The utility function of (2) is defined as:
wherein a is i The calculation processing size of the unmanned aerial vehicle is represented as H i (t) price per unit of task, b i Representing the price of the unmanned aerial vehicle unit computing resource, and u representing the price of the unmanned aerial vehicle unit cost, then, taking the maximized unmanned aerial vehicle utility function as an optimization target to be represented as follows:
C 2 ：E f (t)+E h (t)+E c (t)<RE(t-1),
C 3 :a k ∈(0,1),a i,k ∈(0,1),
C 5 :0≤f i,k ≤f max ,
wherein C1 represents that the total energy consumption of the unmanned aerial vehicle in the process of calculating unloading can not exceed the maximum energy E of the battery of the unmanned aerial vehicle d (ii) a C2 represents that the calculated unloading total energy consumption of the unmanned aerial vehicle at the current moment is less than the residual energy RE (t-1) of the unmanned aerial vehicle battery at the last moment; c3, representing a hovering decision of the unmanned aerial vehicle and an unloading decision of the ground intelligent terminal i; c4 represents that the ground intelligent terminal i can only unload the task to one unmanned aerial vehicle; each unmanned aerial vehicle can only provide calculation unloading service for one ground intelligent terminal at the same time; c5 represents that the computing resource distributed to the ground intelligent terminal i by the unmanned aerial vehicle cannot exceed the maximum computing resource f of the unmanned aerial vehicle max (ii) a C6 represents that the task time delay calculated by the intelligent terminal i does not exceed the maximum tolerance time delay.
5. The intelligent unloading method for the computational tasks under the mapping error condition of the twin network of unmanned aerial vehicles according to claim 1, wherein the step of constructing the optimization target into the Markov decision process by using a competitive near-end strategy optimization method comprises the following steps: defining a state space, an action space and a reward function,
state space: at decision time T, T ∈ {0,1, \8230;, T }, the twin state space of the local base station is defined as:
wherein
Representing drone position, RE (t) representing drone residual energy, w i (t) represents the position of the ground intelligent terminal i, R i (t) represents the calculation task of the ground intelligent terminal i, h i,k (t) represents the channel gain between the ground-based intelligent terminal I and the drone hovering at the kth suspension point, where I ∈ {1,2, \8230;, I }, K ∈ {0,1,.., K }
An action space: at decision time t, the action space that optimizes the utility of the drone is represented as:
t ＝{a k ,a i,k ,f i,k } (13)
wherein a is k Represents the decision of the drone to select the kth suspension point at the time t, a i,k Express an offloading decision of the ground intelligent terminal i, f i,k The unmanned aerial vehicle distributes computing resources of a ground intelligent terminal i;
the reward function: in the process of executing the near-end strategy optimization algorithm, whether the constraints C1-C6 are established or not needs to be checked, so that the following instant rewarding function r is designed t ：
If the constraints C1-C6 cannot be met, indicating that the performance of the current optimization decision is poor, in this case, setting the instant reward to 0 to avoid invalid decisions; in the algorithm training process, the local base station BS obtains an initial state from a maintained twin model, generates a random strategy, obtains a new state and an action transfer sequence through interaction with the twin model, and simultaneously adjusts respective Actor-critical network parameters to obtain an optimal strategy; the loss function of the Actor network parameter update procedure is expressed as:
wherein the probability ratio
Indicating the magnitude of the update of the network parameter,
the value of an action is reflected by the representation advantage function, the epsilon is the clipping rate, and the aim of the clip function is to compare the probability ratio r t (theta) is limited to the interval [1- ∈,1+ ∈ ]]In the formula (1-3), when the merit function is
When r is t Stopping updating when (theta) exceeds 1 +. Epsilon
When r is t (theta) less than 1-epsilon stops the update in order to prevent the updated network from being too different from the old network, thus achieving a small stride update.
6. The intelligent unloading method for the calculation tasks under the unmanned plane twin network mapping error condition as claimed in claim 5, wherein the definition of the merit function is as follows:
δ t ＝r t +V(s t+1 )-V(s t ) (16)
wherein the content of the first and second substances,
representing the dominance function, δ t Represents a single step time error, gamma e (0, 1) represents a discount coefficient, V(s) t+1 ) Represents a state s t+1 Estimated value of (c), V(s) t ) Represents a state s t The estimated value of (c).
7. The intelligent unloading method for the calculation tasks under the unmanned aerial vehicle twin network mapping error condition as claimed in claim 5, wherein the Critic network calculates the state value by using the action value output by the sub-network, and the specific expression is as follows:
V(s t )＝Q(s t ,a) T P(s t ,a) (17)
wherein, P(s) t ,a)＝[p 1 ,p 1 ,…p K ]Respectively taking action p 1 ,p 1 ,…p K Probability of (a), P(s) t A) is derived from the operator network output, while the Critic network will be delta t The mean square error of (2) is used as a loss function, and the gradient descent method is used for minimizing the loss function, so that the network parameters are updated.
8. The utility model provides a calculation task intelligence uninstallation system under unmanned aerial vehicle twin network mapping error condition which characterized in that, this system includes: the intelligent unloading method program for the calculation task under the unmanned aerial vehicle twin network mapping error condition is executed by the processor to realize the following steps:
s1: constructing digital twin models of the intelligent terminal and the unmanned aerial vehicle, and simulating the running states of the intelligent terminal and the unmanned aerial vehicle in a base station service area by using the twin models to generate twin data;
s2: constructing a calculation unloading and resource allocation strategy, and taking a maximized unmanned aerial vehicle utility function as an optimization target;
s3: constructing an optimization target into a Markov decision process, optimizing a competitive near-end strategy by using twin data, and solving the Markov decision process by using the optimized competitive near-end strategy;
s4: and (4) making a flight track, calculating unloading and distributing resources by using the decision action output in the step (S3).
9. The system for intelligently unloading the calculation tasks under the condition of the unmanned aerial vehicle twin network mapping errors according to claim 8, wherein the expression of a digital twin model of an intelligent terminal is as follows:
DT i ＝{M i ,D i ,s i (t),s i (t+1)}
wherein M is i Respectively representing the behavior model of the intelligent terminal, s i (t) respectively representing the real-time status of the intelligent terminal, s i And (t + 1) represents the update state of the intelligent terminal.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium includes a program of a method for intelligently offloading computing tasks in case of unmanned aerial vehicle twin network mapping errors, and when the program of the method for intelligently offloading computing tasks in case of unmanned aerial vehicle twin network mapping errors is executed by a processor, the steps of the method for intelligently offloading computing tasks in case of unmanned aerial vehicle twin network mapping errors are implemented according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211311838.3A CN115659803A (en) | 2022-10-25 | 2022-10-25 | Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211311838.3A CN115659803A (en) | 2022-10-25 | 2022-10-25 | Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115659803A true CN115659803A (en) | 2023-01-31 |
Family
ID=84991257
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211311838.3A Pending CN115659803A (en) | 2022-10-25 | 2022-10-25 | Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115659803A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116643581A (en) * | 2023-04-17 | 2023-08-25 | 重庆控环科技集团有限公司 | Communication unmanned aerial vehicle path planning and bandwidth allocation method considering power consumption faults |
| CN116860447A (en) * | 2023-07-10 | 2023-10-10 | 中国电信股份有限公司技术创新中心 | Task caching method, device, system, equipment and medium |
| CN118075783A (en) * | 2024-04-18 | 2024-05-24 | 中国人民解放军32806部队 | Robust communication method and system based on base station multi-user wireless communication ad hoc network |
| US12027055B1 (en) * | 2023-04-21 | 2024-07-02 | Nanjing University Of Posts And Telecommunications | Unmanned aerial vehicle assisted task offloading and resource allocation method based on service caching |
-
2022
- 2022-10-25 CN CN202211311838.3A patent/CN115659803A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116643581A (en) * | 2023-04-17 | 2023-08-25 | 重庆控环科技集团有限公司 | Communication unmanned aerial vehicle path planning and bandwidth allocation method considering power consumption faults |
| US12027055B1 (en) * | 2023-04-21 | 2024-07-02 | Nanjing University Of Posts And Telecommunications | Unmanned aerial vehicle assisted task offloading and resource allocation method based on service caching |
| CN116860447A (en) * | 2023-07-10 | 2023-10-10 | 中国电信股份有限公司技术创新中心 | Task caching method, device, system, equipment and medium |
| CN118075783A (en) * | 2024-04-18 | 2024-05-24 | 中国人民解放军32806部队 | Robust communication method and system based on base station multi-user wireless communication ad hoc network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112351503B (en) | Task prediction-based multi-unmanned aerial vehicle auxiliary edge computing resource allocation method | |
| CN111800828B (en) | Mobile edge computing resource allocation method for ultra-dense network | |
| CN115659803A (en) | Intelligent unloading method for computing tasks under unmanned aerial vehicle twin network mapping error condition | |
| CN114051254B (en) | Green cloud edge collaborative computing unloading method based on star-ground fusion network | |
| CN112929849B (en) | Reliable vehicle-mounted edge calculation unloading method based on reinforcement learning | |
| CN111915142A (en) | Unmanned aerial vehicle auxiliary resource allocation method based on deep reinforcement learning | |
| Ebrahim et al. | A deep learning approach for task offloading in multi-UAV aided mobile edge computing | |
| CN115065678A (en) | Multi-intelligent-device task unloading decision method based on deep reinforcement learning | |
| CN114626298A (en) | State updating method for efficient caching and task unloading in unmanned aerial vehicle-assisted Internet of vehicles | |
| CN116321293A (en) | Edge computing unloading and resource allocation method based on multi-agent reinforcement learning | |
| CN114025359B (en) | Resource allocation and calculation unloading method, system, equipment and medium based on deep reinforcement learning | |
| CN113821346B (en) | Edge computing unloading and resource management method based on deep reinforcement learning | |
| CN114219074A (en) | Wireless communication network resource allocation algorithm dynamically adjusted according to requirements | |
| CN111930435B (en) | Task unloading decision method based on PD-BPSO technology | |
| CN116737391A (en) | Edge computing cooperation method based on mixing strategy in federal mode | |
| CN114745386B (en) | Neural network segmentation and unloading method in multi-user edge intelligent scene | |
| CN116886158A (en) | DDPG-based star-ground fusion network mobile edge computing resource allocation method | |
| Si et al. | UAV-assisted Semantic Communication with Hybrid Action Reinforcement Learning | |
| Yuan et al. | Joint Multi-Ground-User Edge Caching Resource Allocation for Cache-Enabled High-Low-Altitude-Platforms Integrated Network | |
| CN115955711A (en) | Air-ground 6G network resource allocation method oriented to optimal energy efficiency | |
| CN112969157B (en) | Network load balancing method for unmanned aerial vehicle | |
| CN115967430A (en) | Cost-optimal air-ground network task unloading method based on deep reinforcement learning | |
| CN115766478A (en) | Unloading method of air-ground cooperative edge computing server | |
| CN114513814A (en) | Edge network computing resource dynamic optimization method based on unmanned aerial vehicle auxiliary node | |
| Basharat et al. | Latency and Cost Minimization for Task Offloading with Energy Harvesting in UAV-MEC Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |