CN109803292A - A method of the mobile edge calculations of more secondary user's based on intensified learning - Google Patents
A method of the mobile edge calculations of more secondary user's based on intensified learning Download PDFInfo
- Publication number
- CN109803292A CN109803292A CN201811597091.6A CN201811597091A CN109803292A CN 109803292 A CN109803292 A CN 109803292A CN 201811597091 A CN201811597091 A CN 201811597091A CN 109803292 A CN109803292 A CN 109803292A
- Authority
- CN
- China
- Prior art keywords
- secondary user
- value
- edge server
- server
- edge
- 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.)
- Granted
Links
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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The method of the mobile edge calculations of more secondary user's based on intensified learning that the invention discloses a kind of, this method is based on priority scan method under Dyna structure, suitable for frequency spectrum resource anxiety, time delay and the high mobile edge calculations wireless communications environment of horsepower requirements belong to wireless communication field.This method mainly includes four steps: the selection of main user progress Edge Server first;Secondly secondary user's propose to occupy Edge Server application to control centre;Then control centre's processing application, and allocations of edge server, secondary user's carry out part and calculate unloading or entirely local calculating;Finally calculate the utility of each secondary user's.Intensified learning is applied to mobile edge calculations cordless communication network by this method, and combine the model-free of Q study and the advantage of priority scan method preferentially updated, while guaranteeing system overall utility performance, the demand of each secondary user's time delay and energy consumption is met, the utilization rate of resource is improved.
Description
Technical field
The present invention relates to wireless communication fields, are to be related to a kind of more secondary user's based on intensified learning more specifically
Mobile edge calculations method is a kind of nervous for frequency spectrum resource, time delay and the high mobile edge calculations channel radio of horsepower requirements
Believe the method for environment.
Background technique
It is only applicable to deterministic environment using traditional priority scan method, ideal can not be obtained for circumstances not known
As a result.And the method learnt based on traditional model-free Q, it will do it the update of all state action values, repeatedly calculate all shapes
State action value, the consuming time is longer, can not quickly obtain selection Edge Server and determine mobile edge unloading task amount most
Dominant strategy.In order to improve the frequency spectrum resource utilization rate of whole system, researcher proposes some based on model hypothesis and known logical
Believe the methods of frequency spectrum share and the frequency spectrum resource configuration of environment, but these methods can not take into account each secondary user's effectiveness,
With the convergence rate of Edge Server selection optimal policy.
Summary of the invention
Present invention aim to address traditional mobile edge calculations methods cannot take into account user's time delay horsepower requirements and be
Unite utility, optimal policy convergence rate the problem of, provide based under Dyna structure in intensified learning priority scan it is more
The mobile edge calculations method of secondary user's.
To achieve the above object, technical solution provided by the invention is as follows:
The mobile edge calculations method of more secondary user's based on intensified learning, which is characterized in that the exchange method packet
Include the following steps:
S1, system initialization parameter determine the number N of main userP, the number N of secondary user'sS, of Edge Server
Several and control node number NM, the transmission power P of secondary user's, the task amount of secondary user's is Task, each is secondary
The channel capacity C of channel between user and Edge Server, communication urgency Em is zero;Initial method parameter, edge service
The state of channel corresponding to deviceIt is initially vacant, value zero, initialization Q value is zero, learning rate
For α, discount factor δ, pri function value is zero, priority threshold value θ, and priority query is sky, starts iteration;
S2, main user select to occupy Edge Server MP, the state value of the server is 1;
S3, secondary user's i propose to occupy the application M of Edge Server resource according to ε-greedy strategyi1, and determine and calculate
The task amount x of unloading;
S4, control centre handle the application of each secondary user's, and are its allocations of edge server;
S5, the secondary user's for obtaining Edge Server resource realize that unloading calculates, and do not obtain the secondary user's for occupying qualification,
Then carry out entirely local calculating;
S6, the current effectiveness of each secondary user's is calculated as return, and the Edge Server that will be returned and be successfully connected immediately
It is updated in models of priority;
S7, more new traffic urgency, and Q value and pri function are updated, if pri function is higher than threshold θ, by this
State and selection are added in priority query, and update corresponding Q value according to priority orders;
S8, judge whether to meet stopping criterion for iteration, if satisfied, it is average then to calculate each secondary user's after entire method executes
Effectiveness;If not satisfied, the S2 that then gos to step.
Preferably, the step S4 is specifically included:
S41, control centre handle the application of each secondary user's;
If channel status corresponding to S42, Edge Server is 1, apply for that the secondary user's of the server are unavailable
The right to use jumps to step S44;
If S43, an Edge Server are only applied once, which is accounted for by applied secondary user's
With, i.e., successfully occupy original application server;
If S44, Edge Server are applied twice or more, to be arranged according to the communication urgency of each secondary user's
Sequence, the higher preferential right to use for obtaining Edge Server of communication urgency, that is, occupy non-original application server resource;
S45, the secondary user's for not obtaining the Edge Server right to use, then be probabilistically assigned the edge service not being applied
Device guarantees that all Edge Servers are occupied or secondary user's obtain server resource;Edge Server if it exists
It has been taken that, part secondary user's do not obtain server resource, then secondary user's can only carry out entirely local calculating.
Preferably, the step S6 is specifically included:
S61, each secondary user's utility include two parts, mainly have the delay of calculating and the energy consumption of calculating, with effect
It is inversely proportional with property;
In the value of utility of the secondary user's of S62, progress local computing, include local computing delay and local computing energy consumption;
Carry out calculating section unloading secondary user's value of utility in, comprising local computing delay, local computing energy consumption, unload time delay and
Unload energy consumption.
Preferably, the step S7 is specifically included:
S71, more new traffic degree are mainly determined according to whether secondary user's successfully obtain Edge Server resource, if secondary
User successfully occupies original application server, then it communicates urgency and remains unchanged;If non-original application side only can be used in secondary user's
Edge server, then Em+1;If secondary user's do not obtain the resource of server, it is only capable of entirely local calculating, then its Em+2;
S72, update Q value are the following discount learnt with learning rate α to be returned the error between current Q value, i.e.,
It predicts error, is superimposed on current Q value;Updating pri function is to take prediction error and current priority functional value most
Big value is as new pri function value.
Compared with prior art, the beneficial effects of the present invention are:
1, the mobile edge calculations method of more secondary user's disclosed by the invention based on intensified learning is compared to major part
For traditional mobile edge calculations method, it is suitable for the environment of time-varying, has both and promotes effectiveness, the advantage of fast convergence rate, greatly
It is big to improve resource utilization ratio, secondary user's delay can be met while meeting main user resources demand and is wanted with energy consumption
It asks, is suitable for frequency spectrum resource anxiety, the high mobile edge calculations wireless environment of time delay horsepower requirements.
2, the mobile edge calculations method of more secondary user's disclosed by the invention based on intensified learning, compared to major part
Traditional Q study Edge Server and randomly selected spectrum resource allocation method, method disclosed by the invention can be with faster
Speed convergence to optimal policy.
Detailed description of the invention
The invention will be further described with example with reference to the accompanying drawing.
Fig. 1 is the process step of the mobile edge calculations method of more secondary user's proposed by the present invention based on intensified learning
Figure;
The process step figure of control centre's processing secondary user's application in Fig. 2 present invention;
The process step figure of secondary user's value of utility is calculated in Fig. 3 present invention;
Fig. 4 is the process step figure of more new traffic urgency, Q function and pri function in the present invention;
Fig. 5 (a) is the effectiveness comparison diagram based on simple Q study and the method for the present invention of secondary user's 1;
Fig. 5 (b) is the effectiveness comparison diagram based on simple Q study and the method for the present invention of secondary user's 2;
Fig. 5 (c) is the energy consumption comparison figure based on simple Q study and the method for the present invention of secondary user's 1;
Fig. 5 (d) is the energy consumption comparison figure based on simple Q study and the method for the present invention of secondary user's 2;
Fig. 5 (e) is the time delay comparison diagram based on simple Q study and the method for the present invention of secondary user's 1;
Fig. 5 (f) is the time delay comparison diagram based on simple Q study and the method for the present invention of secondary user's 2.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer and more explicit, right as follows in conjunction with drawings and embodiments
The present invention is further described.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and do not have to
It is of the invention in limiting.
Embodiment one
The present embodiment devises a kind of mobile edge calculations method of more secondary user's based on intensified learning.In the present invention
The process flow of exchange method includes:
S1, system initialization parameter determine the number N of main userP, the number N of secondary user'sS, of Edge Server
Several and control node number NM, the transmission power P of secondary user's, the task amount of secondary user's is Task, each is secondary
The channel capacity C of channel between user and Edge Server, communication urgency Em is zero;Initial method parameter, edge service
The state of channel corresponding to deviceIt is initially vacant, value zero, initialization Q value is zero, learning rate
For α, discount factor δ, pri function value is zero, priority threshold value θ, and priority query is sky, starts iteration;
S2, main user select to occupy Edge Server MP, the state value of the server is 1;
S3, secondary user's i propose to occupy the application M of Edge Server resource according to ε-greedy strategyi1, and determine and calculate
The task amount x of unloading;
S4, control centre handle the application of each secondary user's, and are its allocations of edge server;
S41, the secondary user's for not obtaining the Edge Server right to use, then be probabilistically assigned the edge service not being applied
Device guarantees that all Edge Servers are occupied or secondary user's obtain server resource;Edge Server if it exists
It has been taken that, part secondary user's do not obtain server resource, then secondary user's can only carry out entirely local calculating.
S42, the secondary user's for obtaining Edge Server resource realize that unloading calculates, and do not obtain the secondary of occupancy qualification and use
Family then carries out entirely local calculating;
S6, the current effectiveness of each secondary user's is calculated as return, and the Edge Server that will be returned and be successfully connected immediately
It is updated in models of priority;In the value of utility for carrying out the secondary user's of local computing, include local computing delay and local meter
Calculate energy consumption;In the value of utility for carrying out the secondary user's of calculating section unloading, comprising local computing delay, local computing energy consumption, unload
Carry time delay and unloading energy consumption.
S7, more new traffic urgency, if secondary user's successfully occupy original application server, communication urgency is kept not
Become;If non-original application Edge Server, E only can be used in secondary user'sm+1;If secondary user's do not obtain the money of server
Source is only capable of entirely local calculating, then its Em+2;And update Q value, by following discount learn with learning rate α return and
Error between current Q value, i.e. prediction error, are superimposed on current Q value;And take prediction error and current priority functional value
Maximum value as new pri function value, if pri function is higher than threshold θ, priority is added in this state and selection
In queue, and corresponding Q value is updated according to priority orders;
S8, setting 1000 time slot of iteration and 500 independent experiments.Judge whether to meet stopping criterion for iteration, if satisfied, then
Calculate each secondary user's average utility after entire method executes;If not satisfied, the S2 that then gos to step.
The key of this movement edge calculations method is that user is allow to select mobile edge using intensified learning method
Changeable environment is adapted to during server.Wherein, using the priority scanning algorithm under the Dyna structure in intensified learning,
Learn in conjunction with the higher state of preferential update priority and Q of priority scan method excellent with environmental interaction acquisition experience
Gesture obtains the selection strategy of optimal mobile Edge Server in the case where meeting the requirement of user's time delay and energy consumption.
Embodiment two
In terms of the present embodiment combination Figure of description 1 to Fig. 5 and the mobile edge for specifically containing two secondary user's by one
Embodiment is calculated the mobile edge calculations method proposed by the present invention based on intensified learning is once described in detail.
Consider that system model is as follows: during mobile edge calculations, 1 main user, 2 secondary user's, selection
Using 3 Edge Servers, unloading task is carried out.The task amount of secondary user's 1 is Task1, the task amount of secondary user's 2 is
Task2, the channel capacity between two secondary user's and three Edge Servers is Matrix C, and communication urgency is zero.Edge clothes
The corresponding channel status of business device is zero, i.e., unoccupied.Q value is zero, learning rate 0.8, discount factor 0.02, preferentially
Grade functional value zero, priority threshold value 0.15, and priority query are sky, start iteration.
In each time slot, main user occupies Edge Server first, makes the state 1 of its respective channels, and secondary is used
Family proposes to use the application of Edge Server using ε-greedy strategy, and determines the amount of calculating task unloading.Control centre's processing
The application of each secondary user's, and allocations of edge server guarantee that all Edge Servers are occupied or secondary user's
Obtain server resource;Edge Server has taken if it exists, and part secondary user's do not obtain server resource, then secondary
User can only carry out entirely local calculating.
If carrying out local computing, local time delay and energy consumption are calculated;If carrying out calculating unloading, two parts are calculated,
The time delay and energy consumption that time delay energy consumption, unloading including local computing calculate.And update value of utility, Q value and pri function value.
In each time slot, two secondary user's are influenced by the selection of other side's Edge Server mutually, and target is maximum
Change value of utility.The termination condition of iteration is to carry out 300 time slots altogether, 700 independent experiments.
The simulation result of Fig. 5 (a) shows in terms of value of utility, proposed by the present invention based on strong for secondary user's 1
The mobile edge calculations method that chemistry is practised is than the fast convergence rate about 50% that is learnt based on simple Q.The simulation result of Fig. 5 (b)
Illustrate, for secondary user's 2, fast convergence rate about 30% of the method for the present invention on value of utility than being learnt based on simple Q.By
Shown in Fig. 5 (c) and Fig. 5 (d), method proposed by the present invention all learns the energy consumption of two secondary user's than being based on simple Q
With better performance.Fig. 5 (e) and Fig. 5 (f) show that in terms of calculation delay, method proposed by the present invention has better property
Energy.
To sum up, method proposed by the present invention, in terms of value of utility, calculating energy consumption and calculation delay, compared to based on simple
Q study, guarantee optimal value it is constant and combine two secondary user's performance in the case where, have faster convergence speed
Degree.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (4)
1. a kind of mobile edge calculations method of more secondary user's based on intensified learning, which is characterized in that the calculation method
Include the following steps:
S1, system initialization parameter determine the number N of main userP, the number N of secondary user'sS, the number of Edge Server with
And the number N of control nodeM, the transmission power P of secondary user's, the task amount of secondary user's is Task, each secondary user's
The channel capacity C of channel between Edge Server, communication urgency Em is zero;Initial method parameter, Edge Server institute
The state of corresponding channelIt is initially vacant, value zero, initialization Q value is zero, learning rate α,
Discount factor is δ, and pri function value is zero, priority threshold value θ, and priority query is sky, starts iteration;
S2, main user select to occupy Edge Server MP, the state value of the server is 1;
S3, secondary user's i propose to occupy the application M of Edge Server resource according to ε-greedy strategyi1, and determine and calculate unloading
Task amount x;
S4, control centre handle the application of each secondary user's, and are its allocations of edge server;
S5, obtain Edge Server resource secondary user's realize unloading calculate, do not obtain occupy qualification secondary user's, then into
The entirely local calculating of row;
S6, the current effectiveness of each secondary user's of calculating are updated as return immediately, and by the Edge Server returned and be successfully connected
Into models of priority;
S7, more new traffic urgency, and Q value and pri function are updated, if pri function is higher than threshold θ, by this state
It is added in priority query with selection, and updates corresponding Q value according to priority orders;
S8, judge whether to meet stopping criterion for iteration, averagely be imitated if satisfied, then calculating each secondary user's after entire method executes
With;If not satisfied, the S2 that then gos to step.
2. a kind of mobile edge calculations method of more secondary user's based on intensified learning according to claim 1, special
Sign is that the step S4 is specifically included:
S41, control centre handle the application of each secondary user's;
If channel status corresponding to S42, Edge Server is 1, apply for the unavailable use of the secondary user's of the server
Power, jumps to step S44;
If S43, an Edge Server are only applied once, which is occupied by applied secondary user's, i.e.,
Success occupies original application server;
If S44, Edge Server are applied twice or more, to be ranked up according to the communication urgency of each secondary user's,
The higher preferential right to use for obtaining Edge Server of urgency is communicated, that is, occupies non-original application server resource;
S45, the secondary user's for not obtaining the Edge Server right to use are then probabilistically assigned the Edge Server not being applied, protect
Demonstrate,prove that all Edge Servers are occupied or secondary user's obtain server resource;Edge Server has accounted for if it exists
Full, part secondary user's do not obtain server resource, then secondary user's can only carry out entirely local calculating.
3. a kind of mobile edge calculations method of more secondary user's based on intensified learning according to claim 1, special
Sign is that the step S6 is specifically included:
S61, each secondary user's utility include two parts, mainly have the delay of calculating and the energy consumption of calculating, with utility
It is inversely proportional;
In the value of utility of the secondary user's of S62, progress local computing, include local computing delay and local computing energy consumption;It carries out
In the value of utility of the secondary user's of calculating section unloading, include local computing delay, local computing energy consumption, unloading time delay and unloading
Energy consumption.
4. a kind of mobile edge calculations method of more secondary user's based on intensified learning according to claim 1, special
Sign is that the step S7 is specifically included:
S71, more new traffic degree are mainly determined according to whether secondary user's successfully obtain Edge Server resource, if secondary user's
Success occupies original application server, then it communicates urgency and remains unchanged;If only non-original application edge can be used to take for secondary user's
It is engaged in device, then Em+1;If secondary user's do not obtain the resource of server, it is only capable of entirely local calculating, then its Em+2;
S72, update Q value are by the error between the following discount learnt with learning rate α return and current Q value, that is, to predict
Error is superimposed on current Q value;Updating pri function is the maximum value for taking prediction error and current priority functional value
As new pri function value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811597091.6A CN109803292B (en) | 2018-12-26 | 2018-12-26 | Multi-level user moving edge calculation method based on reinforcement learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811597091.6A CN109803292B (en) | 2018-12-26 | 2018-12-26 | Multi-level user moving edge calculation method based on reinforcement learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109803292A true CN109803292A (en) | 2019-05-24 |
CN109803292B CN109803292B (en) | 2022-03-04 |
Family
ID=66557603
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811597091.6A Active CN109803292B (en) | 2018-12-26 | 2018-12-26 | Multi-level user moving edge calculation method based on reinforcement learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109803292B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110347495A (en) * | 2019-07-24 | 2019-10-18 | 张�成 | A kind of task immigration method carrying out mobile edge calculations using deeply study |
CN110798849A (en) * | 2019-10-10 | 2020-02-14 | 西北工业大学 | Computing resource allocation and task unloading method for ultra-dense network edge computing |
CN111400031A (en) * | 2020-03-01 | 2020-07-10 | 南京大学 | Value function-based reinforcement learning method for processing unit deployment |
CN112512056A (en) * | 2020-11-14 | 2021-03-16 | 北京工业大学 | Multi-objective optimization calculation unloading method in mobile edge calculation network |
CN113709201A (en) * | 2020-05-22 | 2021-11-26 | 华为技术有限公司 | Method and communication device for computing offloading |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103248693A (en) * | 2013-05-03 | 2013-08-14 | 东南大学 | Large-scale self-adaptive composite service optimization method based on multi-agent reinforced learning |
US20170261949A1 (en) * | 2016-03-11 | 2017-09-14 | University Of Chicago | Apparatus and method for optimizing quantifiable behavior in configurable devices and systems |
CN108632860A (en) * | 2018-04-17 | 2018-10-09 | 浙江工业大学 | A kind of mobile edge calculations rate maximization approach based on deeply study |
CN108924938A (en) * | 2018-08-27 | 2018-11-30 | 南昌大学 | A kind of resource allocation methods for wireless charging edge calculations network query function efficiency |
-
2018
- 2018-12-26 CN CN201811597091.6A patent/CN109803292B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103248693A (en) * | 2013-05-03 | 2013-08-14 | 东南大学 | Large-scale self-adaptive composite service optimization method based on multi-agent reinforced learning |
US20170261949A1 (en) * | 2016-03-11 | 2017-09-14 | University Of Chicago | Apparatus and method for optimizing quantifiable behavior in configurable devices and systems |
CN108632860A (en) * | 2018-04-17 | 2018-10-09 | 浙江工业大学 | A kind of mobile edge calculations rate maximization approach based on deeply study |
CN108924938A (en) * | 2018-08-27 | 2018-11-30 | 南昌大学 | A kind of resource allocation methods for wireless charging edge calculations network query function efficiency |
Non-Patent Citations (3)
Title |
---|
LIANG XIAO 等: "《Security_in_Mobile_Edge_Caching_with_Reinforcement_Learning》", 《IEEE WIRELESS COMMUNICATIONS》 * |
XIAOZHEN LU 等: "《Learning-Based_Rogue_Edge_Detection_in_VANETs_with_Ambient_Radio_Signals》", 《2018 IEEE INTERNATIONAL CONFERENCE ON COMMUNICATIONS (ICC)》 * |
ZHENYU ZHOU 等: "《An_Air-Ground_Integration_Approach_for_Mobile_Edge_Computing_in_IoT》", 《IEEE COMMUNICATIONS MAGAZINE》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110347495A (en) * | 2019-07-24 | 2019-10-18 | 张�成 | A kind of task immigration method carrying out mobile edge calculations using deeply study |
CN110347495B (en) * | 2019-07-24 | 2023-04-28 | 张�成 | Task migration method for performing mobile edge calculation by using deep reinforcement learning |
CN110798849A (en) * | 2019-10-10 | 2020-02-14 | 西北工业大学 | Computing resource allocation and task unloading method for ultra-dense network edge computing |
CN111400031A (en) * | 2020-03-01 | 2020-07-10 | 南京大学 | Value function-based reinforcement learning method for processing unit deployment |
CN111400031B (en) * | 2020-03-01 | 2023-08-22 | 南京大学 | Value function-based reinforcement learning method for processing unit deployment |
CN113709201A (en) * | 2020-05-22 | 2021-11-26 | 华为技术有限公司 | Method and communication device for computing offloading |
CN112512056A (en) * | 2020-11-14 | 2021-03-16 | 北京工业大学 | Multi-objective optimization calculation unloading method in mobile edge calculation network |
CN112512056B (en) * | 2020-11-14 | 2022-10-18 | 北京工业大学 | Multi-objective optimization calculation unloading method in mobile edge calculation network |
Also Published As
Publication number | Publication date |
---|---|
CN109803292B (en) | 2022-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109803292A (en) | A method of the mobile edge calculations of more secondary user's based on intensified learning | |
Zeng et al. | Resource management at the network edge: A deep reinforcement learning approach | |
Wang et al. | Intelligent cognitive radio in 5G: AI-based hierarchical cognitive cellular networks | |
Sun et al. | Autonomous resource slicing for virtualized vehicular networks with D2D communications based on deep reinforcement learning | |
Nassar et al. | Reinforcement learning for adaptive resource allocation in fog RAN for IoT with heterogeneous latency requirements | |
CN111093203B (en) | Service function chain low-cost intelligent deployment method based on environment perception | |
Jošilo et al. | Joint wireless and edge computing resource management with dynamic network slice selection | |
Huynh et al. | Joint computational offloading and data-content caching in NOMA-MEC networks | |
Zhou et al. | Learning from peers: Deep transfer reinforcement learning for joint radio and cache resource allocation in 5G RAN slicing | |
CN112672382B (en) | Hybrid collaborative computing unloading method and device, electronic equipment and storage medium | |
Jiang et al. | A q-learning based method for energy-efficient computation offloading in mobile edge computing | |
Balakrishnan et al. | Deep reinforcement learning based traffic-and channel-aware OFDMA resource allocation | |
CN116541106B (en) | Computing task unloading method, computing device and storage medium | |
Wang et al. | Multi-agent reinforcement learning-based user pairing in multi-carrier NOMA systems | |
Wang et al. | Online convex optimization for efficient and robust inter-slice radio resource management | |
CN113810910B (en) | Deep reinforcement learning-based dynamic spectrum sharing method between 4G and 5G networks | |
Gao et al. | Com-DDPG: A multiagent reinforcement learning-based offloading strategy for mobile edge computing | |
Geng et al. | A Reinforcement learning framework for vehicular network routing under peak and average constraints | |
Mafuta et al. | Decentralized resource allocation-based multiagent deep learning in vehicular network | |
Mei et al. | Semi-decentralized network slicing for reliable V2V service provisioning: A model-free deep reinforcement learning approach | |
CN110392377A (en) | A kind of 5G super-intensive networking resources distribution method and device | |
Albonda et al. | Analysis of RAN slicing for Cellular V2X and mobile broadband services based on reinforcement learning | |
CN110224861A (en) | The implementation method of adaptive dynamic heterogeneous network selection policies based on study | |
Qin et al. | Energy‐aware resource allocation for OFDMA wireless networks with hybrid energy supplies | |
Cui et al. | Resource-Efficient DNN Training and Inference for Heterogeneous Edge Intelligence in 6G |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |