CN114828018A - Multi-user mobile edge computing unloading method based on depth certainty strategy gradient - Google Patents

Multi-user mobile edge computing unloading method based on depth certainty strategy gradient Download PDF

Info

Publication number
CN114828018A
CN114828018A CN202210325855.6A CN202210325855A CN114828018A CN 114828018 A CN114828018 A CN 114828018A CN 202210325855 A CN202210325855 A CN 202210325855A CN 114828018 A CN114828018 A CN 114828018A
Authority
CN
China
Prior art keywords
user
task
mec server
network
delay
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
Application number
CN202210325855.6A
Other languages
Chinese (zh)
Inventor
王睿
史敏燕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Original Assignee
Tongji University
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 Tongji University filed Critical Tongji University
Priority to CN202210325855.6A priority Critical patent/CN114828018A/en
Publication of CN114828018A publication Critical patent/CN114828018A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/02—Resource partitioning among network components, e.g. reuse partitioning
    • H04W16/10—Dynamic resource partitioning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/53—Allocation or scheduling criteria for wireless resources based on regulatory allocation policies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/535—Allocation or scheduling criteria for wireless resources based on resource usage policies

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention belongs to the field of wireless communication, and provides a mobile edge server task unloading method based on a depth certainty strategy gradient. Under the condition that a plurality of users exist in the system, the invention achieves the purpose of optimizing the task processing time delay of the system to the minimum by reasonably distributing system resources. The invention adopts a depth certainty strategy gradient algorithm to solve the power distribution problem of processing the tasks at the local end and the mobile edge server end, thereby reducing the task processing time delay in the system. The method models the calculation unloading problem into the problem of minimum maximum calculation time delay of the whole system under the time variation of a channel, simultaneously considers the energy constraints of an edge server and a user, constructs an optimization target conforming to the model of the reinforcement learning algorithm, and solves the calculation unloading strategy by using a depth certainty strategy gradient algorithm.

Description

Multi-user mobile edge computing unloading method based on depth certainty strategy gradient
Technical Field
The invention belongs to the field of wireless communication, and relates to a task unloading method for a mobile edge server, in particular to a task unloading method for the mobile edge server based on a depth certainty strategy gradient.
Background
The mobile edge computing reduces time delay and energy consumption brought by data transmission by deploying the base station in the environment at a position closer to the terminal equipment, so as to meet the requirement of processing a large amount of mobile user data in a short time and ensure the use experience and service quality of users. Computation offload in a mobile edge computing network requires that offload policies be followed. The current main task unloading optimization algorithms include a game algorithm, a convex optimization algorithm, a dynamic programming algorithm and the like. However, the mobile edge computing system has high real-time performance, and parameters such as channel states and task volumes in the system change with the environment, and the traditional optimization method for solving the problems often needs to be iterated for many times and cannot adapt to the high real-time environment of the system. In addition, along with diversification of mobile edge computing unloading scenes, the computing unloading strategy also involves multiple links such as server selection, task allocation and the like, and challenges are brought to solving of the optimal computing unloading strategy. The deep reinforcement learning algorithm is a machine learning algorithm which can carry out self-learning according to feedback given by the environment and dynamically adjust the strategy, and the neural network fitting reward function is used for accelerating the convergence speed, so that the deep reinforcement learning algorithm can more effectively solve the problem of computation unloading in a complex scene compared with the traditional optimization method.
Disclosure of Invention
The invention aims to solve the task and resource allocation problem of a user and an MEC server (mobile edge computing server) under a multi-user computing unloading model.
By adopting a partial unloading mode, a user can simultaneously carry out task calculation locally and can also transmit the task to the mobile edge server for calculation. The connection between the user and the MEC server is variable, considering that there are multiple users in the system. For the problem, the connection scene is subdivided into a one-to-one mode (scene 1) and a one-to-many mode (scene 2), and the two modes are respectively modeled systematically and a resource allocation scheme is proposed.
The method models the calculation unloading problem into the problem of minimizing the maximum calculation delay of the whole system under the condition of channel time variation, simultaneously considers the energy constraints of an edge server and a user, constructs an optimization target conforming to a reinforcement learning algorithm model, and solves an optimal calculation unloading scheme by using a depth certainty strategy gradient.
A multi-user mobile edge computing unloading method based on a depth certainty strategy gradient comprises the following steps:
step 1, initializing a current value network Q, a current strategy network mu, a target value network Q ', a target strategy network mu', wherein the parameters of the current network are theta respectively Q 、θ μ The parameter of the target network is theta Q' ←θ Q ，θ μ' ←θ μ 。
Step 2, initializing an experience pool B;
step 3, at the time t, the current task amount C to be processed of the user is calculated k (t) set of channel gains h for different users and mobile edge servers at the previous time k (t-1)＝{h k1 (t-1),h k2 (t-1),...,h kM (t-1) } as the current system state s k (t)＝[h k (t-1),C k (t)]；
And 4, at the moment t, selecting the action a (t) by the user according to the observed system state quantity, selecting the action with the maximum Q value according to the probability of 1-epsilon, and randomly selecting other actions according to the probability of epsilon. Wherein the greedy factor epsilon is a constant of 0-1. The action set includes task allocation and power allocation:
Figure BDA0003573453820000031
α k (t) local task weight for user k, α k C k Representing the amount of tasks a user processes locally, (1-alpha) k )C k Representing the amount of tasks a user offloads to the MEC server.
Figure BDA0003573453820000032
Representing the local processing power of user k,
Figure BDA0003573453820000033
represents the data transmission power of user k,
Figure BDA0003573453820000034
representing the computational power that MEC server m allocates to the tasks offloaded by user k.
Step 5, judging the connection scene of the action according to the action selected by the system, and respectively calculating different connectionsAnd receiving the maximum task processing time delay in the scene. The task processed locally is a k C k The local processing delay of user k can be calculated as:
Figure BDA0003573453820000035
the calculation of the unloading delay comprises two parts: and the transmission delay of the task during unloading and the processing delay of the task at the MEC server are completed only after the task is processed on the MEC server, and the whole task unloading process is finished. According to the amount of unloaded tasks (1-a) k )C k Then the transmission delay for user k to offload the task to MEC server m is:
Figure BDA0003573453820000036
the processing delay of the task unloaded by the user k is calculated by the MEC server m as follows:
Figure BDA0003573453820000037
local computation and task offloading are performed simultaneously, so that the current task computation delay of user k is the maximum of the local processing delay and the computation offloading delay:
Figure BDA0003573453820000041
in scenario 1, in one timeslot, one MEC server can be connected only by one user, and the connections between different users and the MEC server are independent and do not affect each other. In scenario 2, different users may connect to the same MEC server at the same time, which means that limited resources of one MEC server are provided for multiple users, and there is a competitive relationship between users. The reward function of the system is:
Figure BDA0003573453820000042
after the action is executed, calculating the current reward value according to the formula (5), and observing the next state;
step 6, the experience(s) at the time t is compared t ,a t ,r t ,s t+1 ) Putting the mixture into an experience pool B;
step 7, randomly extracting samples from the experience pool B to train a current value network and a current strategy network;
and 8, calculating a target Q value according to a formula (6):
y i ＝r i +γQ′(S i+1 ,μ′(S i+1 |θ μ′ )|θ Q′ ) (6)
and 9, minimizing a loss function of the current value network:
Figure BDA0003573453820000043
step 10, updating the parameter theta of the current strategy network by adopting a random gradient descent method μ ：
Figure BDA0003573453820000044
Wherein J is a strategy objective function;
Figure BDA0003573453820000051
is a policy gradient;
Figure BDA0003573453820000052
is the gradient of the Q function;
Figure BDA0003573453820000053
as a function of the movement μ(s) t |θ μ )]Of the gradient of (a).
And 11, updating parameters of the target value network and parameters of the target strategy network:
θ Q' ←τθ Q +(1-τ)θ Q' (9)
θ μ' ←τθ μ +(1-τ)θ μ' (10)
wherein, theta μ′ Is a policy network parameter; theta Q′ A parameter that is a value network; τ is an update weight factor;
and 12, according to the environmental change, the user repeats the steps 3-11 until the strategy is converged, and at the moment, the user learns the optimal calculation unloading scheme.
Further, in step 5, the user local computation delay is derived from the user local computation rate, and the local computation rate of the user k is:
Figure BDA0003573453820000054
wherein D is k Is the number of CPU cycles required for user k to complete a task bit (meaning that the size of the task is one bit),
Figure BDA0003573453820000055
is the calculated power allocated locally by user k,
Figure BDA0003573453820000056
is the CPU frequency of user k at time t, k l Is the effective switched capacitance (CPU number/sec) depending on the user equipment chip architecture,
further, in step 5, the task processing delay of the MEC server is derived from the computing rate of the MEC server and the task transmission rate, and the computing rate of the MEC server m is:
Figure BDA0003573453820000057
wherein D is m Is the number of CPU cycles, k, required for the MEC server m to complete a task bit m Is an effective switched capacitor (CPU number/sec) depending on the architecture of the MEC server chip)，
Figure BDA0003573453820000061
Is the computing power that MEC server m allocates to the task that user k offloaded at time t.
The communication rate between user k and MEC server m is:
Figure BDA0003573453820000062
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003573453820000063
represents the data transmission power when the user k carries out task unloading at the time t, W represents the system bandwidth, N 0 Representing the variance of gaussian white noise.
Advantageous effects
The invention achieves the purpose of optimizing the minimum time delay of system task processing by reasonably distributing system resources. The power distribution problem of processing tasks at a local end and a mobile edge server end is solved by adopting a depth deterministic strategy gradient algorithm, so that the task processing time delay in the system is reduced. The mobile edge calculation unloading model provided by the invention considers the mutual influence of a plurality of users, subdivides a connection scene into a one-to-one mode and a one-to-many mode, and respectively carries out system modeling on the two modes and provides a resource allocation scheme. Modeling the calculation unloading problem as the problem of minimum maximum calculation time delay of the whole system under the time variation of a channel, simultaneously considering the energy constraints of an edge server and a user, constructing an optimization target conforming to the model of the reinforcement learning algorithm, and solving the calculation unloading strategy by using a depth certainty strategy gradient algorithm.
Drawings
FIG. 1 is a block diagram of a model framework for a multi-user mobile edge computing offload system
FIG. 2 is a flowchart of a moving edge calculation offload optimization algorithm based on deep reinforcement learning
Detailed Description
The technical solutions provided in the present application will be further described with reference to the following specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent in conjunction with the following description.
The method models the calculation unloading problem into the problem of minimizing the maximum calculation time delay of the whole system under the time variation of the channel, considers the energy constraints of the edge server and the user, constructs an optimization target in accordance with the model of the reinforcement learning algorithm, and solves the optimal calculation unloading scheme by using the gradient of the depth certainty strategy.
The method comprises the following steps:
1) initializing a current value network Q, a current strategy network mu, a target value network Q ', a target strategy network mu', and setting the parameters of the current network to theta Q 、θ μ The parameter of the target network is theta Q' ←θ Q ，θ μ' ←θ μ . The four networks have the same structure and comprise five layers, namely an input layer, three hidden layers and an output layer, wherein the three hidden layers respectively comprise 200 neurons, 150 neurons and 50 neurons. The learning rates of the value network and the strategy network are 0.0001 and 0.001 respectively;
2) initializing a memory playback unit B of 2 × 10 5 ；
3) The number of users in the system is 2, and the number of MEC servers is 2. At the time t, the current task amount C to be processed of the user is calculated k (t) set of channel gains h for different users and mobile edge servers at the previous time k (t-1)＝{h k1 (t-1),h k2 (t-1),...,h kM (t-1) } as the current system state s k (t)＝[h k (t-1),C k (t)]；
4) At time t, the user selects action a (t) according to the observed system state quantity, selects the action with the maximum Q value with the probability of 0.9, and randomly selects other actions with the probability of 0.1. The action set includes task allocation and power allocation:
Figure BDA0003573453820000081
α k (t) local task weight for user k, α k C k Representing the amount of tasks a user processes locally, (1-alpha) k )C k For indicatingThe amount of tasks a customer offloads to the MEC server.
Figure BDA0003573453820000082
Representing the local processing power of user k,
Figure BDA0003573453820000083
represents the data transmission power of user k,
Figure BDA0003573453820000084
representing the computational power that MEC server m allocates to the tasks offloaded by user k.
5) And judging the connection scene to which the action belongs according to the action selected by the system, and respectively calculating the maximum task processing time delay under different connection scenes. The task processed locally is a k C k The local processing delay of user k can be calculated as:
Figure BDA0003573453820000085
the calculation of the unloading delay comprises two parts: and the transmission delay of the task during unloading and the processing delay of the task at the MEC server are completed only after the task is processed on the MEC server, and the whole task unloading process is finished. According to the amount of unloaded tasks (1-a) k )C k Then the transmission delay for user k to offload the task to MEC server m is:
Figure BDA0003573453820000086
the processing delay of the task unloaded by the user k is calculated by the MEC server m as follows:
Figure BDA0003573453820000087
local computation and task offloading are performed simultaneously, so that the current task computation delay of user k is the maximum of the local processing delay and the computation offloading delay:
Figure BDA0003573453820000091
in scenario 1, in one timeslot, one MEC server can be connected only by one user, and the connections between different users and the MEC server are independent and do not affect each other. In scenario 2, different users may connect to the same MEC server at the same time, which means that limited resources of one MEC server are provided for multiple users, and there is a competitive relationship between users. The reward function of the system is:
Figure BDA0003573453820000092
after the action is executed, calculating the current reward value according to the formula (5), and observing the next state;
6) the experience(s) at time t t ,a t ,r t ,s t+1 ) Putting the mixture into an experience pool B;
7) randomly extracting samples from the experience pool B to train a current value network and a current strategy network;
8) the target Q value is calculated according to equation (6):
y i ＝r i +γQ'(S i+1 ,μ'(S i+1 |θ μ' )|θ Q' ) (6)
9) minimizing the loss function of the current value network:
Figure BDA0003573453820000093
10) updating parameter theta of current strategy network by adopting random gradient descent method μ ：
Figure BDA0003573453820000094
11) Updating parameters of the target value network and parameters of the target strategy network:
θ Q' ←τθ Q +(1-τ)θ Q' (9)
θ μ' ←τθ μ +(1-τ)θ μ' (10)
and according to the environment change, the user repeats the steps 3) to 11) until the strategy is converged, and at the moment, the user learns the optimal calculation unloading scheme.

Claims (3)

1. A multi-user mobile edge computing unloading method based on a depth certainty strategy gradient is characterized by comprising the following steps:
step 1, initializing a current value network Q, a current strategy network mu, a target value network Q ', a target strategy network mu', wherein the parameters of the current network are theta respectively Q 、θ μ The parameter of the target network is theta Q' ←θ Q ，θ μ' ←θ μ ；
Step 2, initializing an experience pool B;
step 3, at the time t, the current task amount C to be processed of the user is calculated k (t) set of channel gains h for different users and mobile edge servers at the previous time k (t-1)＝{h k1 (t-1),h k2 (t-1),...,h kM (t-1) } as the current system state s k (t)＝[h k (t-1),C k (t)]；
Step 4, at the moment t, selecting an action a (t) by a user according to the observed system state quantity, selecting the action with the maximum Q value according to the probability of 1-epsilon, and randomly selecting other actions according to the probability of epsilon, wherein a greedy factor epsilon is a constant of 0-1; the action set includes task allocation and power allocation:
Figure FDA0003573453810000011
wherein alpha is k (t) local task weight for user k, α k C k Indicating that the user is at homeTask amount of theory, (1-alpha) k )C k Representing the amount of tasks a user offloads to the MEC server;
Figure FDA0003573453810000012
representing the local processing power of user k,
Figure FDA0003573453810000013
represents the data transmission power of user k,
Figure FDA0003573453810000014
represents the computational power that the MEC server m allocates to the tasks offloaded by user k;
step 5, judging the connection scene to which the action belongs according to the action selected by the system, and respectively calculating the maximum task processing time delay under different connection scenes; the task processed locally is a k C k The local processing delay of user k can be calculated as:
Figure FDA0003573453810000021
the calculation of the unloading delay comprises two parts: the transmission delay of the task during unloading and the processing delay of the task in the MEC server are completed, and the whole task unloading process is only completed after the task is processed on the MEC server; according to the amount of unloaded tasks (1-a) k )C k Then the transmission delay for user k to offload the task to MEC server m is:
Figure FDA0003573453810000022
the processing delay of the task unloaded by the user k is calculated by the MEC server m as follows:
Figure FDA0003573453810000023
local computation and task offloading are performed simultaneously, so that the current task computation delay of user k is the maximum of the local processing delay and the computation offloading delay:
Figure FDA0003573453810000024
in a scenario 1, in a time slot, one MEC server can only be connected by one user, and the connections between different users and the MEC server are independent and do not affect each other;
in scenario 2, different users can connect to the same MEC server at the same time, which means that limited resources of one MEC server are supplied to a plurality of users, and a competitive relationship exists between users; the reward function of the system is:
Figure FDA0003573453810000025
after the action is executed, calculating the current reward value according to the formula (5), and observing the next state;
step 6, the experience(s) at the time t is compared t ,a t ,r t ,s t+1 ) Putting the mixture into an experience pool B;
step 7, randomly extracting samples from the experience pool B to train a current value network and a current strategy network;
and 8, calculating a target Q value according to a formula (6):
y i ＝r i +γQ′(S i+1 ,μ′(S i+1 |θ μ′ )|θ Q′ )
(6)
and 9, minimizing a loss function of the current value network:
Figure FDA0003573453810000031
step 10, updating the current by adopting a random gradient descent methodParameter θ of policy network μ ：
Figure FDA0003573453810000032
Wherein J is a strategy objective function;
Figure FDA0003573453810000033
is a policy gradient;
Figure FDA0003573453810000034
is the gradient of the Q function;
Figure FDA0003573453810000035
as a function of the movement μ(s) t |θ μ )]A gradient of (a);
and 11, updating parameters of the target value network and parameters of the target strategy network:
θ Q' ←τθ Q +(1-τ)θ Q' (9)
θ μ' ←τθ μ +(1-τ)θ μ' (10)
wherein, theta μ′ Is a policy network parameter; theta Q′ A parameter of a value network; τ is an update weight factor;
and 12, according to the environmental change, the user repeats the steps 3-11 until the strategy is converged, and at the moment, the user learns the optimal calculation unloading scheme.
2. The method for offloading computation of multi-user moving edge based on depth deterministic policy gradient of claim 1, wherein in step 5, the user local computation delay is derived from the user local computation rate, and the local computation rate of user k is:
Figure FDA0003573453810000041
wherein D is k Is the number of CPU cycles required for user k to complete a task bit (meaning that the size of the task is one bit),
Figure FDA0003573453810000042
is the calculated power allocated locally by user k,
Figure FDA0003573453810000043
is the CPU frequency of user k at time t, k l Is the effective switched capacitance (CPU number/sec) depending on the user equipment chip architecture.
3. The method according to claim 1, wherein in step 5, the task processing delay of the MEC server is derived from a computing rate of the MEC server and a task transmission rate, and the computing rate of the MEC server m is:
Figure FDA0003573453810000044
wherein D is m Is the number of CPU cycles, k, required for the MEC server m to complete a task bit m Is dependent on the effective switched capacitance (CPU number/sec) of the MEC server chip architecture,
Figure FDA0003573453810000045
the computing power of the tasks unloaded by the user k is distributed by the MEC server m at the moment t;
the communication rate between user k and MEC server m is:
Figure FDA0003573453810000046
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003573453810000047
represents the data transmission power when the user k carries out task unloading at the time t, W represents the system bandwidth, N 0 Representing the variance of gaussian white noise.
CN202210325855.6A 2022-03-30 2022-03-30 Multi-user mobile edge computing unloading method based on depth certainty strategy gradient Pending CN114828018A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210325855.6A CN114828018A (en) 2022-03-30 2022-03-30 Multi-user mobile edge computing unloading method based on depth certainty strategy gradient

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210325855.6A CN114828018A (en) 2022-03-30 2022-03-30 Multi-user mobile edge computing unloading method based on depth certainty strategy gradient

Publications (1)

Publication Number Publication Date
CN114828018A true CN114828018A (en) 2022-07-29

Family

ID=82531996

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210325855.6A Pending CN114828018A (en) 2022-03-30 2022-03-30 Multi-user mobile edge computing unloading method based on depth certainty strategy gradient

Country Status (1)

Country Link
CN (1) CN114828018A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115366099A (en) * 2022-08-18 2022-11-22 江苏科技大学 Mechanical arm depth certainty strategy gradient training method based on forward kinematics
CN116156566A (en) * 2023-04-18 2023-05-23 南京邮电大学 Multi-access edge computing non-independent task unloading method
CN116709428A (en) * 2023-08-04 2023-09-05 华东交通大学 Calculation unloading method and system based on mobile edge calculation

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115366099A (en) * 2022-08-18 2022-11-22 江苏科技大学 Mechanical arm depth certainty strategy gradient training method based on forward kinematics
CN115366099B (en) * 2022-08-18 2024-05-28 江苏科技大学 Mechanical arm depth deterministic strategy gradient training method based on forward kinematics
CN116156566A (en) * 2023-04-18 2023-05-23 南京邮电大学 Multi-access edge computing non-independent task unloading method
CN116156566B (en) * 2023-04-18 2023-07-25 南京邮电大学 Multi-access edge computing non-independent task unloading method
CN116709428A (en) * 2023-08-04 2023-09-05 华东交通大学 Calculation unloading method and system based on mobile edge calculation
CN116709428B (en) * 2023-08-04 2023-11-24 华东交通大学 Calculation unloading method and system based on mobile edge calculation

Similar Documents

Publication Publication Date Title
CN111800828B (en) Mobile edge computing resource allocation method for ultra-dense network
CN112367353B (en) Mobile edge computing unloading method based on multi-agent reinforcement learning
CN109862610B (en) D2D user resource allocation method based on deep reinforcement learning DDPG algorithm
CN114828018A (en) Multi-user mobile edge computing unloading method based on depth certainty strategy gradient
CN107766135B (en) Task allocation method based on particle swarm optimization and simulated annealing optimization in moving cloud
CN111405569A (en) Calculation unloading and resource allocation method and device based on deep reinforcement learning
CN111629380B (en) Dynamic resource allocation method for high concurrency multi-service industrial 5G network
WO2021227508A1 (en) Deep reinforcement learning-based industrial 5g dynamic multi-priority multi-access method
CN109756578B (en) Low-delay task scheduling method for dynamic fog computing network
CN111130911B (en) Calculation unloading method based on mobile edge calculation
CN114138373B (en) Edge computing task unloading method based on reinforcement learning
CN114340016B (en) Power grid edge calculation unloading distribution method and system
CN110519849B (en) Communication and computing resource joint allocation method for mobile edge computing
CN113467952A (en) Distributed federated learning collaborative computing method and system
Ma et al. A strategic game for task offloading among capacitated UAV-mounted cloudlets
CN114885420A (en) User grouping and resource allocation method and device in NOMA-MEC system
CN113626104B (en) Multi-objective optimization unloading strategy based on deep reinforcement learning under edge cloud architecture
CN112650581A (en) Cloud-side cooperative task scheduling method for intelligent building
CN116456493A (en) D2D user resource allocation method and storage medium based on deep reinforcement learning algorithm
CN112788605A (en) Edge computing resource scheduling method and system based on double-delay depth certainty strategy
CN116744311B (en) User group spectrum access method based on PER-DDQN
CN116489708B (en) Meta universe oriented cloud edge end collaborative mobile edge computing task unloading method
Cheng et al. Efficient resource allocation for NOMA-MEC system in ultra-dense network: A mean field game approach
CN116893861A (en) Multi-agent cooperative dependency task unloading method based on space-ground cooperative edge calculation
CN117098189A (en) Computing unloading and resource allocation method based on GAT hybrid action multi-agent reinforcement learning

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