CN111340277A - Popularity prediction model and method based on federal learning in fog wireless access network - Google Patents
Popularity prediction model and method based on federal learning in fog wireless access network Download PDFInfo
- Publication number
- CN111340277A CN111340277A CN202010102397.0A CN202010102397A CN111340277A CN 111340277 A CN111340277 A CN 111340277A CN 202010102397 A CN202010102397 A CN 202010102397A CN 111340277 A CN111340277 A CN 111340277A
- Authority
- CN
- China
- Prior art keywords
- scene
- content
- prediction model
- request
- popularity prediction
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 238000005457 optimization Methods 0.000 claims description 24
- 238000012549 training Methods 0.000 claims description 24
- 238000004422 calculation algorithm Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000010354 integration Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 abstract description 6
- 230000005540 biological transmission Effects 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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
Abstract
The invention discloses a content popularity prediction model construction method and a content popularity prediction model prediction method based on federal learning in a fog wireless access network, which enable edge nodes in the fog wireless access network to construct a local popularity prediction model according to user preference in a node coverage range, accurately predict the popularity of content, and integrate a global model on the basis of each local model, thereby reducing the calculation complexity and the transmission cost.
Description
Technical Field
The invention relates to a popularity prediction model based on federal learning and a construction method thereof, belonging to the technical field of content popularity prediction in mobile communication.
Background
With the rapid development of smart devices, wireless networks face countless challenges, especially the data traffic pressure of fronthaul wireless links. To address this problem, the fog radio access network (F-RAN) has become an effective solution to alleviate the traffic burden on the fronthaul link by caching popular content in the fog access point (F-AP). In a fog radio access network, fog access points with limited caching and computing resources are densely deployed at the edge of the network to meet the user's request. Due to the limitations of cache capacity, F-APs need to accurately predict future content popularity in order to prefetch the most popular content during off-peak traffic, improving cache efficiency.
Conventional caching strategies, such as the least recently used caching strategy and the least frequently used caching strategy, are widely used in wired networks. However, due to the limited coverage and storage space of the edge nodes in the wireless network, the above conventional caching strategies cannot directly predict the content popularity in advance, and may suffer from severe performance degradation in the wireless network. Therefore, these conventional caching strategies have very limited applicability in wireless networks. Currently, there have been many research efforts aimed at improving caching efficiency by predicting popularity of content. If the future content popularity can be accurately predicted, the cache configuration efficiency of the edge node can be greatly improved, so that the cache hit rate is improved to the maximum extent.
Federal learning is an emerging artificial intelligence base technology. The method is originally used for solving the problem that an android mobile phone terminal user updates a model locally, and the design goal is to carry out efficient machine learning among multiple parties or multiple computing nodes on the premise of guaranteeing information safety during big data exchange, protecting terminal data and personal data privacy and guaranteeing legal compliance.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problems and the defects in the prior art, the popularity prediction model based on the federal learning and the construction method thereof are provided, the popularity of the content in the coverage range of the edge node in the fog wireless access network can be accurately predicted, so that the cache hit rate is improved, the communication cost is low, and the calculated amount is small.
The technical scheme is as follows: a method for constructing a content popularity prediction model based on federal learning in a fog wireless access network comprises the following steps:
s000: according to the historical request information of each user, establishing an objective function of a preference model corresponding to each user, and solving to obtain user preference by taking the minimized objective function as an optimization objective; the historical request information comprises characteristic information of the request content and a user request information tag;
s100: according to the user scene information, a scene space is constructed, and all the obtained scene spaces are subjected to self-adaptive space division to obtain a scene subspace;
s200: according to the similarity between the user preference and the characteristic information of the request content, calculating to obtain the request probability of the user for the request content and the average request probability of all the users in each scene subspace;
s300: taking the average request probability as input, establishing a local popularity prediction model, and solving to obtain the local popularity prediction model by taking the minimum mean square error as an optimization target;
s400: and integrating the local popularity prediction model by adopting a distributed computing method to obtain a global popularity prediction model based on a federated learning framework.
Further, the objective function of the preference model in S000 is represented as:
in the formula ：
denoted as user unHistory request information ofiDenoted as request content ciCharacteristic information of (a), yn,i∈ {0,1} denotes a user request information tag, wnDenoted as user preferences.
Further, to minimize F (w)n) Solving the optimization target by an FTRL-Proximal algorithm to obtain the user preference wn。
Further, the S100 specifically operates as follows:
s110: one user establishes a group of scene information, quantizes each group of scene information and normalizes the scene information to a [0,1] interval;
s120: according to the normalized scene information, constructing an initial scene space theta corresponding to the normalized scene information0Is [0,1]]DD is the dimension of the scene information, theta0Division level l of0＝0；
S130: determining an initial scene space Θ0Whether the number of users included in the list satisfies
wherein ,r1 and r2If the parameters are preset parameters and the parameters are satisfied, the initial scene space theta is determined0Is divided equally to obtain 2DEach scene subspace is divided into 1 on the basis of the division level; if not, not performing the dividing operation;
s140: satisfy the number of users
All scene subspaces of (2) are respectively equally spaced and divided intoDA scene subspace,/j＝li+1, formula (O)iFor the ith scene subspace,/iIs thetaiThe division level of (1);
s150: repeating S140 until the obtained number of users in all the scene subspaces is not satisfied
Then, the segmentation process is finished, and the obtained scene subspace is recorded as theta1,Θ2,...,Θs,...ΘS。
Further, the user preference w is calculated according to the following formulanCharacteristic information χ of request contentiThe similarity between the users u and the users unProbability of request for the content:
the probability value range is [0,1 ].
Further, the average request probability for all users within each scene subspace is calculated according to the following formula:
wherein ,xs,iIn the scene subspace theta for the requested contentsAverage request probability within, Num (Θ)s) Is thetasTotal number of users in (1).
Further, the local popularity prediction model is expressed as:
wherein ,
to request content ciPredicted popularity value, a ═ a1,a2,…,as,…,aS]TPredicting model parameters of the model for local popularity, wherein parameter asRepresenting a scene subspace ΘsActivity level of user, xi＝[x1,i,x2,i,…,xS,i]TThe average request probability of all scenario subspaces under the coverage range of each access node is collected;
establishing an optimization model by taking the minimum mean square error as an optimization target:
wherein ,piIs content ciThe actual value of the popularity of the product,
is a set of training examples on a single node, L is
Total number of training samples;
and solving the optimization model to obtain a local popularity prediction model.
Further, solving the formula (7) by adopting an SVRG algorithm to obtain a local popularity prediction model.
Further, the S400 specifically operates as follows:
s410: in the scenario of K access node integration, according to equation (7), the optimization model may be established as:
wherein 
For the set of all training examples in K access nodes, L*Is composed of
The number of training examples;
s420: the mean square error in each node can be expressed as:
wherein ,
is a set of training examples in the mth node, LmIs composed of
The number of training examples;
s430: rewriting the objective function in equation (14) to obtain a new optimization model:
s440: and combining the DANE algorithm with the SVRG algorithm, and solving a formula (17) by using a distributed calculation method by using a federated learning framework to obtain a global popularity prediction model.
The invention also discloses a content popularity prediction method, which adopts a global popularity prediction model to predict the content popularity in the coverage range of the edge node in the fog wireless access network to obtain a prediction result; and caching the corresponding popular content in the fog access point according to the prediction result.
Has the advantages that: the invention has the following advantages:
(1) according to the method, the relevance between the user preference and the content characteristics is researched by learning a potential user preference model, so that the popularity of new content without statistical data can be predicted;
(2) according to the method, the contextual information of the users is fully utilized, and the users are clustered through self-adaptive contextual space segmentation, so that the number of parameters in a popularity prediction model is reduced, and the calculation amount is greatly reduced;
(3) the distributed computation is carried out by utilizing the federated learning framework, so that the computation complexity on a single node is reduced, and the communication cost is reduced;
(4) the invention can enable the edge node in the fog wireless access network to construct a local popularity prediction model according to the user preference in the node coverage range, accurately predict the popularity of the content, and integrate a global model on the basis of each local model, thereby reducing the computational complexity and the transmission cost.
Drawings
FIG. 1 is a schematic diagram of a fog radio access network scenario contemplated by the present invention;
FIG. 2 is a flow chart of a popularity prediction model and a construction method thereof according to the present invention;
FIG. 3 is a detailed flow chart of a popularity prediction model construction method proposed by the present invention;
fig. 4 is an explanatory diagram of an adaptive scene space division method.
FIG. 5 is a graph of simulation results of predictions using the popularity prediction model obtained in the present invention and compared with the conventional prediction model (the horizontal axis is the number of contents participating in the prediction, and the vertical axis is the root mean square error between the predicted value and the actual value);
fig. 6 is a simulation result graph (the horizontal axis is cache capacity, and the vertical axis is cache hit rate) in which caching is performed according to a prediction result and the obtained cache hit rate is compared with a conventional algorithm;
fig. 7 is a simulation result diagram of the convergence rate of the algorithm (the horizontal axis represents the number of iterations of the algorithm, and the vertical axis represents the objective function value) compared with the distributed algorithm based on federal learning and the centralized algorithm employed in the present invention.
Detailed Description
The technical solution of the present invention will be further explained with reference to the accompanying drawings and examples.
The popularity prediction model based on the federal learning and the construction method thereof comprise the following steps:
s000: establishing a preference model of each user according to historical request information data in user equipment and the characteristics of contents, and independently training and learning user preferences; the specific operation is as follows:
s010: recording user unHistory request information of
Including the requested content ciCharacteristic information χ ofiAnd user request information tag yn,i∈ {0,1}, if user u is in the history request informationnRequest content ci，y n,i1, otherwise, yn,i＝0；
S020: utilizing a sigmoid function to approximate the corresponding relation between the characteristic information of the request content and the information tag of the user request:
s030: establishing user unRequest probability function of (1):
i.e. user unThe characteristic information of the requested content is χiThe probability of requesting the content;
s040: building user u by using enough request information as training samplenLikelihood function of (d):
s050: establishing a cross entropy loss function as an objective function of a user preference model:
the above equation is a negative log-likelihood function.
S060: to minimize F (w)n) For optimizing The target, The user preference w is obtained by solving through The Follow The (Proximally) regulated leader (FTRL-Proximal) algorithmnAnd each user performs independent training according to the S010-S060 line, and no information interaction exists among the users.
S100: each user sends the self context information to a corresponding access node after screening and encryption, the node constructs a context space according to the collected context information, sets parameters and performs self-adaptive space segmentation, thereby realizing the clustering of the users; the specific operation is as follows:
s110: collecting user u by access nodenZeta scene informationn＝[ζn,1,ζn,2,...ζn,D]T∈[0,1]DWhere D is the dimension of the context information, and the context information ζnIs normalized, and common situation information can include age, gender, occupationEtc.;
s120: constructing an initial scene space Θ0Is [0,1]]D，Θ0Division level l of0＝0；
S130: setting a parameter r1 and r2When theta is equal to0Number of users contained in
When, will theta0Are equally spaced apart, theta0Is divided into 2DEach scene subspace, wherein the division level of each subspace is added with 1 on the basis of the division level before the division;
s140: similarly, for the obtained scene subspace ΘiWhen is coming into contact with
When, will thetaiIs equally spaced and divided into 2DA scene subspace, letjFor one of the scene subspaces, then lj＝li+1。
S150: by analogy, when the obtained number of users in all the scene spaces does not meet the segmentation condition, the adaptive scene space segmentation process is ended, and finally all the obtained scene subspaces are marked as theta1,Θ2,...,Θs,...ΘS；
S200: mapping the similarity between the user preference and the content characteristics into the request probability of the user for the content, and taking the average request probability of all users in each segmented scene subspace;
the specific operation is as follows:
s210: mapping function is taken
Characteristic information χ of content to be requestediAnd user preferences wnSubstitution into f (w)n,χi) Formula, calculating the similarity between the two, and obtaining the user unFor the request content ciThe probability of request is in the range of [0,1]]；
S220: for the subspace theta belonging to the same scenesThe request probabilities of the users within are averaged, i.e.:
wherein ,xs,iIs content ciIn the scene subspace ΘsAverage request probability within, Num (Θ)s) Is thetasTotal number of users in (1).
S300: establishing a popularity prediction model by taking the average request probability of the user for the content as input, establishing an optimization problem by using historical data of popularity, and training the popularity prediction model;
the specific operation is as follows:
s310: each node collects the average request probability x of all scenario subspaces under its coveragei=[x1,i,x2,i,...,xS,i]TAs input to a popularity prediction model;
s320: let a = [ a =1,a2,...,as,...,aS]TPredicting model parameters of the model for popularity, wherein parameter asRepresenting scene space ΘsThe activity level of the user, i.e. the contribution to the network traffic;
s330: establishing a local popularity prediction model as shown in the following formula:
wherein ,
is content ciA predicted popularity value;
s340: based on the minimum mean square error criterion, an optimization model is established, so that model parameters are solved:
wherein ,piIs content ciThe actual value of the popularity of the product,
is a set of training examples on a single node, L is
Total number of training samples;
s350: the optimization problem in equation (7) is solved by using the Stochastic Variable Reduced Gradient (SVRG) algorithm.
The specific solving process is as follows:
s351: note the book
vi(a)＝|pi-aTxi|2(8)
Initializing a parameter a, wherein the step length is h, and the iteration number calculated in each round is
S352: calculating an extrinsic cycle gradient value:
s353: setting parameter a*A, randomly draw training examples
The inner loop gradient values are calculated according to the following equation:
g＝▽vi(a*)-(▽vi(a)-▽v(a))， (11)
wherein (▽vi(a)-▽v (a) is for the gradient value ▽ vi(a*) The correction term of (1);
s354: and (3) updating parameter values:
a*＝a*-hg (12)
s355: repeating S353 and S354 to perform
Performing secondary iteration;
s356: and (3) updating parameter values:
a＝a*(13)
s357: repeating S352 to S356 until the objective function value converges;
s400: based on a federated learning framework, a local popularity prediction model is integrated into a global popularity prediction model through a distributed computing method.
The specific operation is as follows:
s410: in the scenario of K access node integration, according to equation (7), the optimization model may be established as:
wherein 
For the set of all training examples in K access nodes, L*Is composed of
The number of training examples;
s420: the mean square error in each node can be expressed as:
wherein ,
is a set of training examples in the mth node, LmIs composed of
The number of training examples;
s430: the objective function in equation (14) is rewritten as:
obtaining a new optimization model:
this indicates that the global optimization is decided by the local optimization;
s440: combining a Distributed Adaptive Newton (DANE) algorithm with a Stochastic VarianceReduced Gradient (SVRG) algorithm, and solving a new optimization model by a distributed computing method by utilizing a framework of federal learning to obtain a global popularity prediction model.
The solving process is as follows:
s441: initializing a parameter a, wherein the step length is h, and the iteration number calculated in each round is
S442: the extrinsic gradient values are calculated as follows:
s443: for all nodes 1 to K, S444 to S446 are performed, respectively;
s444: for the m node, take amA, randomly draw training examples
Calculating an inner loop gradient value according to the following formula;
gm＝▽vi(am)-▽vi(a)+▽v(a) (19)
s445: and (3) updating parameter values:
am＝am-hgm(20)
s446: repeating S444 and S445, proceeding
Performing secondary iteration;
s447: and (3) updating parameter values:
s448: and repeating the steps from S442 to S447 until the objective function value is converged to obtain a global popularity prediction model.
FIG. 5 is a graph of simulation results of predictions using the popularity prediction model obtained in the present invention and compared with conventional prediction models, with the horizontal axis representing the number of contents participating in the predictions and the vertical axis representing the root mean square error between the predicted values and the actual values; FIG. 6 is a graph of simulation results of caching according to the prediction results and comparing the resulting cache hit rate with the conventional algorithm, with cache capacity on the horizontal axis and cache hit rate on the vertical axis; fig. 7 is a simulation result diagram of algorithm convergence rate obtained by comparing the distributed algorithm based on federal learning and the centralized algorithm employed in the present invention, in which the horizontal axis represents the number of iterations of the algorithm and the vertical axis represents the objective function value. Therefore, the invention can enable the edge nodes in the fog wireless access network to construct local popularity prediction models according to the user preference in the node coverage range, accurately predict the popularity of the content, and integrate global models on the basis of each local model, thereby reducing the computational complexity and the transmission cost.
Claims (10)
1. A method for constructing a content popularity prediction model based on federal learning in a fog wireless access network is characterized in that: the method comprises the following steps:
s000: according to the historical request information of each user, establishing an objective function of a preference model corresponding to each user, and solving to obtain user preference by taking the minimized objective function as an optimization objective; the historical request information comprises characteristic information of the request content and a user request information tag;
s100: according to the user scene information, a scene space is constructed, and all the obtained scene spaces are subjected to self-adaptive space division to obtain a scene subspace;
s200: according to the similarity between the user preference and the characteristic information of the request content, calculating to obtain the request probability of the user for the request content and the average request probability of all the users in each scene subspace;
s300: taking the average request probability as input, establishing a local popularity prediction model, and solving to obtain the local popularity prediction model by taking the minimum mean square error as an optimization target;
s400: and integrating the local popularity prediction model by adopting a distributed computing method to obtain a global popularity prediction model based on a federated learning framework.
2. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 1, wherein: the objective function of the preference model in S000 is expressed as:
in the formula ：
denoted as user unHistory request information ofiDenoted as request content ciCharacteristic information of (a), yn,i∈ {0,1} denotes a user request information tag, wnDenoted as user preferences.
3. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 2, wherein: to minimize F (w)n) Solving the optimization target by an FTRL-Proximal algorithm to obtain the user preference wn。
4. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 1, wherein: the S100 specifically operates as follows:
s110: one user establishes a group of scene information, quantizes each group of scene information and normalizes the scene information to a [0,1] interval;
s120: according to the normalized scene information, constructing an initial scene space theta corresponding to the normalized scene information0Is [0,1]]DD is the dimension of the scene information, theta0Division level l of0＝0；
S130: determining an initial scene space Θ0Whether the number of users included in the list satisfies
wherein ,r1 and r2If the parameters are preset parameters and the parameters are satisfied, the initial scene space theta is determined0Is divided equally to obtain 2DEach scene subspace is divided into 1 on the basis of the division level; if not, not performing the dividing operation;
s140: satisfy the number of users
All scene subspaces of (2) are respectively equally spaced and divided intoDA scene subspace,/j＝li+1, formula (O)iFor the ith scene subspace,/iIs thetaiThe division level of (1);
s150: repeating S140 until the obtained number of users in all scene subspacesAll amounts are not satisfied
Then, the segmentation process is finished, and the obtained scene subspace is recorded as theta1,Θ2,...,Θs,...ΘS。
5. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 1, wherein: calculating a user preference w according tonCharacteristic information χ of request contentiThe similarity between the users u and the users unProbability of request for the content:
the probability value range is [0,1 ].
6. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 5, wherein:
the average request probability for all users within each scene subspace is calculated according to:
wherein ,xs,iIn the scene subspace theta for the requested contentsAverage request probability within, Num (Θ)s) Is thetasTotal number of users in (1).
7. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 1, wherein:
the local popularity prediction model is expressed as:
wherein ,
to request content ciPredicted popularity value, a ═ a1,a2,…,as,…,aS]TPredicting model parameters of the model for local popularity, wherein parameter asRepresenting a scene subspace ΘsActivity level of user, xi＝[x1,i,x2,i,...,xS,i]TThe average request probability of all scenario subspaces under the coverage range of each access node is collected;
establishing an optimization model by taking the minimum mean square error as an optimization target:
wherein ,piIs content ciThe actual value of the popularity of the product,
is a set of training examples on a single node, L is
Total number of training samples;
and solving the optimization model to obtain a local popularity prediction model.
8. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 7, wherein: and (5) solving the formula (7) by adopting an SVRG algorithm to obtain a local popularity prediction model.
9. The method for constructing a content popularity prediction model based on federal learning in a fog wireless access network as claimed in claim 7, wherein: the S400 specifically operates as follows:
s410: in the scenario of K access node integration, according to equation (7), the optimization model may be established as:
wherein 
For the set of all training examples in K access nodes, L*Is composed of
The number of training examples;
s420: the mean square error in each node can be expressed as:
wherein ,
is a set of training examples in the mth node, LmIs composed of
The number of training examples;
s430: rewriting the objective function in equation (14) to obtain a new optimization model:
s440: and combining the DANE algorithm with the SVRG algorithm, and solving a formula (17) by using a distributed calculation method by using a federated learning framework to obtain a global popularity prediction model.
10. The method for predicting content popularity based on the method for constructing the content popularity prediction model based on the federal learning in the fog wireless access network as claimed in any one of claims 1 to 9, is characterized in that:
predicting the content popularity in the coverage range of the edge node in the fog wireless access network by adopting a global popularity prediction model to obtain a prediction result;
and caching the corresponding popular content in the fog access point according to the prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010102397.0A CN111340277B (en) | 2020-02-19 | 2020-02-19 | Popularity prediction model and prediction method based on federal learning in fog wireless access network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010102397.0A CN111340277B (en) | 2020-02-19 | 2020-02-19 | Popularity prediction model and prediction method based on federal learning in fog wireless access network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111340277A true CN111340277A (en) | 2020-06-26 |
| CN111340277B CN111340277B (en) | 2023-04-25 |
Family
ID=71185451
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010102397.0A Active CN111340277B (en) | 2020-02-19 | 2020-02-19 | Popularity prediction model and prediction method based on federal learning in fog wireless access network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111340277B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111865826A (en) * | 2020-07-02 | 2020-10-30 | 大连理工大学 | Active content caching method based on federal learning |
| CN112181971A (en) * | 2020-10-27 | 2021-01-05 | 华侨大学 | Edge-based federated learning model cleaning and equipment clustering method, system, equipment and readable storage medium |
| CN113326128A (en) * | 2021-05-28 | 2021-08-31 | 东南大学 | Privacy protection popularity prediction method based on unsupervised loop federal learning in mobile edge computing network |
| CN113381886A (en) * | 2021-06-08 | 2021-09-10 | 东南大学 | Content popularity prediction method based on Bayesian learning in fog wireless access network |
| CN113379066A (en) * | 2021-06-10 | 2021-09-10 | 重庆邮电大学 | Federal learning method based on fog calculation |
| CN113568973A (en) * | 2021-07-21 | 2021-10-29 | 湖南天河国云科技有限公司 | Financial credit investigation data sharing method and device based on block chain and federal learning |
| CN113923128A (en) * | 2021-10-27 | 2022-01-11 | 东南大学 | Intelligent coding caching method based on federal reinforcement learning in fog wireless access network |
| CN113965937A (en) * | 2021-10-27 | 2022-01-21 | 东南大学 | Clustering federal learning-based content popularity prediction method in fog wireless access network |
| CN113992770A (en) * | 2021-10-29 | 2022-01-28 | 东南大学 | Cooperative caching method based on policy-based federal reinforcement learning in fog wireless access network |
| CN114189899A (en) * | 2021-12-10 | 2022-03-15 | 东南大学 | User equipment selection method based on random aggregation beam forming |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109429265A (en) * | 2017-08-29 | 2019-03-05 | 鸿海精密工业股份有限公司 | For controlling the method and device of network flow |
| CN109873869A (en) * | 2019-03-05 | 2019-06-11 | 东南大学 | A kind of edge cache method based on intensified learning in mist wireless access network |
| CN110062421A (en) * | 2019-04-08 | 2019-07-26 | 东南大学 | For the dove colony optimization algorithm in mist wireless access network and based on the cooperation caching method of the algorithm |
-
2020
- 2020-02-19 CN CN202010102397.0A patent/CN111340277B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109429265A (en) * | 2017-08-29 | 2019-03-05 | 鸿海精密工业股份有限公司 | For controlling the method and device of network flow |
| CN109873869A (en) * | 2019-03-05 | 2019-06-11 | 东南大学 | A kind of edge cache method based on intensified learning in mist wireless access network |
| CN110062421A (en) * | 2019-04-08 | 2019-07-26 | 东南大学 | For the dove colony optimization algorithm in mist wireless access network and based on the cooperation caching method of the algorithm |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111865826A (en) * | 2020-07-02 | 2020-10-30 | 大连理工大学 | Active content caching method based on federal learning |
| CN111865826B (en) * | 2020-07-02 | 2022-01-04 | 大连理工大学 | Active content caching method based on federal learning |
| CN112181971A (en) * | 2020-10-27 | 2021-01-05 | 华侨大学 | Edge-based federated learning model cleaning and equipment clustering method, system, equipment and readable storage medium |
| CN113326128A (en) * | 2021-05-28 | 2021-08-31 | 东南大学 | Privacy protection popularity prediction method based on unsupervised loop federal learning in mobile edge computing network |
| CN113381886A (en) * | 2021-06-08 | 2021-09-10 | 东南大学 | Content popularity prediction method based on Bayesian learning in fog wireless access network |
| CN113379066B (en) * | 2021-06-10 | 2022-07-08 | 重庆邮电大学 | Federal learning method based on fog calculation |
| CN113379066A (en) * | 2021-06-10 | 2021-09-10 | 重庆邮电大学 | Federal learning method based on fog calculation |
| CN113568973B (en) * | 2021-07-21 | 2023-11-24 | 湖南天河国云科技有限公司 | Financial credit investigation data sharing method and device based on blockchain and federal learning |
| CN113568973A (en) * | 2021-07-21 | 2021-10-29 | 湖南天河国云科技有限公司 | Financial credit investigation data sharing method and device based on block chain and federal learning |
| CN113965937A (en) * | 2021-10-27 | 2022-01-21 | 东南大学 | Clustering federal learning-based content popularity prediction method in fog wireless access network |
| CN113923128A (en) * | 2021-10-27 | 2022-01-11 | 东南大学 | Intelligent coding caching method based on federal reinforcement learning in fog wireless access network |
| CN113965937B (en) * | 2021-10-27 | 2024-02-13 | 东南大学 | Content popularity prediction method based on clustered federal learning in fog wireless access network |
| CN113923128B (en) * | 2021-10-27 | 2024-02-13 | 东南大学 | Intelligent coding caching method based on federal reinforcement learning in fog wireless access network |
| CN113992770A (en) * | 2021-10-29 | 2022-01-28 | 东南大学 | Cooperative caching method based on policy-based federal reinforcement learning in fog wireless access network |
| CN113992770B (en) * | 2021-10-29 | 2024-02-09 | 东南大学 | Policy-based federal reinforcement learning collaborative caching method in fog wireless access network |
| CN114189899A (en) * | 2021-12-10 | 2022-03-15 | 东南大学 | User equipment selection method based on random aggregation beam forming |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111340277B (en) | 2023-04-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111340277A (en) | Popularity prediction model and method based on federal learning in fog wireless access network | |
| CN112181666A (en) | Method, system, equipment and readable storage medium for equipment evaluation and federal learning importance aggregation based on edge intelligence | |
| CN111901392B (en) | Mobile edge computing-oriented content deployment and distribution method and system | |
| CN111726811B (en) | Slice resource allocation method and system for cognitive wireless network | |
| CN111405569A (en) | Calculation unloading and resource allocation method and device based on deep reinforcement learning | |
| CN112995950B (en) | Resource joint allocation method based on deep reinforcement learning in Internet of vehicles | |
| WO2021253835A1 (en) | Heterogeneous network cache decision-making method based on user preference prediction | |
| CN108833352B (en) | Caching method and system | |
| CN111092823A (en) | Method and system for adaptively adjusting congestion control initial window | |
| CN110856268B (en) | Dynamic multichannel access method for wireless network | |
| CN113485826B (en) | Load balancing method and system for edge server | |
| CN113098714A (en) | Low-delay network slicing method based on deep reinforcement learning | |
| Fragkos et al. | Artificial intelligence enabled distributed edge computing for Internet of Things applications | |
| CN112188503A (en) | Dynamic multichannel access method based on deep reinforcement learning and applied to cellular network | |
| Yang et al. | Deep reinforcement learning based wireless network optimization: A comparative study | |
| CN109787696B (en) | Cognitive radio resource allocation method based on case reasoning and cooperative Q learning | |
| CN112533237B (en) | Network capacity optimization method for supporting large-scale equipment communication in industrial internet | |
| CN116126130A (en) | Task unloading method for trusted edge server selection and energy consumption optimization | |
| CN112291284B (en) | Content pushing method and device and computer readable storage medium | |
| CN113411826B (en) | Edge network equipment caching method based on attention mechanism reinforcement learning | |
| CN113194031B (en) | User clustering method and system combining interference suppression in fog wireless access network | |
| Zhou et al. | Content placement with unknown popularity in fog radio access networks | |
| CN112988275B (en) | Task perception-based mobile edge computing multi-user computing unloading method | |
| CN115129888A (en) | Active content caching method based on network edge knowledge graph | |
| CN114598655A (en) | Mobility load balancing method based on 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |