CN107592656B - Caching method based on base station clustering - Google Patents
Caching method based on base station clustering Download PDFInfo
- Publication number
- CN107592656B CN107592656B CN201710704882.3A CN201710704882A CN107592656B CN 107592656 B CN107592656 B CN 107592656B CN 201710704882 A CN201710704882 A CN 201710704882A CN 107592656 B CN107592656 B CN 107592656B
- Authority
- CN
- China
- Prior art keywords
- base station
- content
- class
- base stations
- clustering
- 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.)
- Active
Links
Abstract
The invention discloses a caching method based on base station clustering, which comprises the steps of firstly collecting and analyzing historical requests of all base station service users under a dense base station network, clustering base stations based on the historical requests, wherein the users served by all base stations in each class have similar interests; meanwhile, the cache content of each base station is decided by combining the collaborative filtering in the field of the recommendation system; by adopting the cluster-based collaborative filtering, the expandability and the data sparsity of the algorithm can be effectively improved. The invention combines the local popularity of the content and the TOPN cooperative filtering system, effectively improves the cache hit rate of the base station, and can effectively solve the contradiction between the limited cache capacity of the base station and the continuously increased mass data, thereby improving the user satisfaction and the network backhaul load.
Description
Technical Field
The invention relates to the technical field of mobile communication systems, in particular to a caching method based on base station clustering.
Background
In order to deal with the challenge to the system capacity brought by the increase of mass data, an effective scheme is to deploy a cache on a base station, and if a user requests content in the cache, the base station directly transmits the content through a wireless link; otherwise it needs to be acquired from the core network via the backhaul link. The active storage of the base station is to store the content in the base station before the request arrives, so that the flow of a return link can be reduced, the flow load in a cellular system is further relieved, and the performance of the system is improved. The invention provides a caching strategy based on base station clustering by analyzing historical requests.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a base station clustering-based cache method, which can greatly improve the cache hit rate, effectively alleviate the backhaul link load, and improve the user satisfaction.
In order to solve the technical problem, the invention provides a caching method based on base station clustering, which comprises the following steps: (1) firstly, the problem of base station clustering is considered; counting the request frequency of each base station service user about the content in the past period, regarding each base station, taking the request frequency of the service user to each content as the characteristic of the base station, clustering the base stations by adopting k-means clustering, wherein the users served by the base stations of each class have similar interests and hobbies, namely the requested content has a plurality of similar places;
(2) the Top N collaborative filtering recommendation system based on the base station predicts the content which is not requested by the user and covered by the base station by utilizing the similarity between the base stations;
(3) and giving each class in the first step, performing targeted caching on the base station by combining the collaborative filtering on each class according to the distribution of the content popularity in the class, and determining the content cached by each base station.
Preferably, the specific clustering of the base stations in step (1) includes the following steps:
(11) according to historical request information of a period of time in the past, a content popularity matrix is obtained through analysis of data by a core networkWherein each element p in the matrixm,fRepresenting the frequency of requests for content f from users served by the base station m, the frequency of requests for content being characteristic of the base station, each row P of the matrix PmA vector representing an F dimension, representing a feature vector of a base station;
(12) randomly selecting k base stations as initial central points of base station clusters, and expressing feature vectors of the k base stations as initial central pointsHere, theSuperscript (1) denotes the first round, i.e. the initial value, and the subscript denotes the ith center point;
(13) from the center point of each class, to minimize the intra-class sum of squares, it is determined to which class the base station belongs as follows:here, theIndicating that the t-th round belongs to the base station set of the i-th class;
(15) repeating (13) and (14) to ciIs less than a given threshold, finally obtaining k class, H1,...,HkEach base station belongs to one of the classes.
Preferably, in the step (2), the intra-class base station-based collaborative filtering specifically includes the following steps:
(21) calculating similarity between base stations within classThe similarity between the intra-class base stations is calculated by the following similarity formula:
base station miAnd base station mjBelong to the same class, T (m)i) And T (m)j) Respectively represent base stations miAnd base station mjThe served users access a collection of content; t (f) represents a set of base stations that have accessed the content f;
(22) from (21), m can be obtainediSet S (m) of closest base stationsiG), then base station miThe served user has an interest level in the content f that has never been requested for a period of time in the past
Where t (f) is the set of base stations that have made requests for content f,is a base station mjThe level of interest in f of the content, here the elements of the content popularity matrix P.
Preferably, the specific caching method in step (3) includes the following steps:
(31) firstly, analyzing the content popularity in each class, namely counting the request contents of all base station service users in the class, and sequencing the request contents according to the access times of the contents from high to low;
(32) buffer capacity of each base station m is Sm(ii) a Eta is the cache capacity S occupied by the content cached through the intra-class popularitymThe content is firstly cached to the base station from high to low according to the intra-class traffic, and before caching the content, whether the total size of the cached content exceeds eta S is checkedmIf yes, giving up the cache;
(33) and for the residual buffer capacity of the base station, performing buffer by a step (22) based on base station cooperative filtering in the class, and performing buffer on the content from high to low according to p (m, f) until the total amount of the buffer content is larger than the buffer capacity.
The invention has the beneficial effects that: the invention provides a caching strategy based on base station clustering by carrying out clustering analysis on the base stations, on one hand, the interest and hobbies of base station service users can be well judged, on the other hand, the complexity of base station collaborative filtering is greatly reduced, and the algorithm performance is improved; the invention combines the local popularity of the content and the TOPN cooperative filtering system, effectively improves the cache hit rate of the base station, and can effectively solve the contradiction between the limited cache capacity of the base station and the continuously increased mass data, thereby improving the user satisfaction and the network backhaul load; compared with the prior art, the method and the device have the advantages that the base stations are clustered, and the machine learning algorithm is introduced into the prediction of the cache content, so that the cache hit rate is greatly improved, the load of a backhaul link is effectively relieved, and the user satisfaction is improved.
Detailed Description
A caching method based on base station clustering comprises the following steps:
(1) firstly, the problem of base station clustering is considered; counting the request frequency of each base station service user about the content in the past period, regarding each base station, taking the request frequency of the service user to each content as the characteristic of the base station, clustering the base stations by adopting k-means clustering, wherein the users served by the base stations of each class have similar interests and hobbies, namely the requested content has a plurality of similar places;
(2) the Top N collaborative filtering recommendation system based on the base station predicts the content which is not requested by the user and covered by the base station by utilizing the similarity between the base stations;
(3) and giving each class in the first step, performing targeted caching on the base station by combining the collaborative filtering on each class according to the distribution of the content popularity in the class, and determining the content cached by each base station.
Preferably, the specific clustering of the base stations in step (1) includes the following steps:
(11) according to historical request information of a period of time in the past, a content popularity matrix is obtained through analysis of data by a core networkWherein each element p in the matrixm,fRepresenting the frequency of requests for content f from users served by the base station m, the frequency of requests for content being characteristic of the base station, each row P of the matrix PmA vector representing an F dimension, representing a feature vector of a base station;
(12) randomly selecting k base stations as initial central points of base station clusters, and expressing feature vectors of the k base stations as initial central pointsHere, theSuperscript (1) denotes the first round, i.e. the initial value, and the subscript denotes the ith center point;
(13) from the center point of each class, to minimize the intra-class sum of squares, it is determined to which class the base station belongs as follows:here, theIndicating that the t-th round belongs to the base station set of the i-th class;
(15) repeating (13) and (14) to ciIs less than a given threshold, finally obtaining k class, H1,...,HkEach base station belongs to one of the classes.
Preferably, in the step (2), the intra-class base station-based collaborative filtering specifically includes the following steps:
(21) calculating similarity between base stations within classThe similarity between the intra-class base stations is calculated by the following similarity formula:
base station miAnd base station mjBelong to the same class, T (m)i) And T (m)j) Respectively represent base stations miAnd base station mjThe served users access a collection of content; t (f) represents a set of base stations that have accessed the content f;
(22) from (21), m can be obtainediSet S (m) of closest base stationsiG), then base station miThe served user has an interest level in the content f that has never been requested for a period of time in the past
Wherein T (f) is of a base station having made a request for the content fIn the collection of the images, the image data is collected,is a base station mjThe level of interest in f of the content, here the elements of the content popularity matrix P.
Preferably, the specific caching method in step (3) includes the following steps:
(31) firstly, analyzing the content popularity in each class, namely counting the request contents of all base station service users in the class, and sequencing the request contents according to the access times of the contents from high to low;
(32) buffer capacity of each base station m is Sm(ii) a Eta is the cache capacity S occupied by the content cached through the intra-class popularitymThe content is firstly cached to the base station from high to low according to the intra-class traffic, and before caching the content, whether the total size of the cached content exceeds eta S is checkedmIf yes, giving up the cache;
(33) and for the residual buffer capacity of the base station, performing buffer by a step (22) based on base station cooperative filtering in the class, and performing buffer on the content from high to low according to p (m, f) until the total amount of the buffer content is larger than the buffer capacity.
Example (b):
network deployment considering M base stationsEach base station is connected to the core network through a backhaul link with a buffer capacity of SmThe content request is collected asThe size of each content is l (f). R (m) serves the set of user request content for base station m, c (m) caches the set of content for base station m. We define the cache hit rate as follows:
the caching method is implemented in the following mode and comprises the following steps:
(1) firstly, the problem of base station clustering is considered, specifically, the frequency of requests of each base station service user about content in the past period is counted. For each base station, the request frequency of the service users to each content is taken as the characteristic of the base station, k-means clustering is adopted to cluster the base stations, and the users served by the base stations of each class have similar interests and hobbies, namely the requested contents have many similar places.
(2) The Top N collaborative filtering recommendation system based on the base station predicts the content which is not requested by the user and covered by the base station by utilizing the similarity between the base stations.
(3) Given each class in the first step, the method performs targeted caching on the base station according to the distribution of content popularity in the class and in combination with performing cooperative filtering on each class, and determines the content cached by each base station.
The step (1) of clustering base stations includes
(11) Analysis of the data by the core network based on historical request information over a period of time. We can derive a content popularity matrixWherein each element p in the matrixm,fRepresenting the frequency of requests for content f by users served by base station m. The request frequency of the content is taken as the characteristic of the base station, and each row P of the matrix PmA vector representing the F dimension represents the feature vector of a base station.
(12) Randomly selecting k base stations as initial central points of base station clusters, wherein the characteristic vectors are expressed as
(13) From the center point of each class, to minimize the intra-class sum of squares, it is determined to which class the base station belongs as follows:
(15) (15) repeating (13), (14) up to ciIs less than a given threshold. Finally obtaining k, H1,...,HkEach base station belongs to one of the classes.
The step (2) is based on the intra-class base station collaborative filtering, and comprises the following specific steps:
(21) calculating similarity between base stations within classThe similarity between the intra-class base stations is calculated by the following similarity formula:
base station miAnd base station mjBelong to the same class, T (m)i) And T (m)j) Respectively represent base stations miBase station mjThe served users access the collection of content. Where t (f) represents the set of base stations that have accessed the content f.
(22) From (21), m can be obtainediSet S (m) of closest base stationsiG), then base station miThe served user has an interest level in the content f that has never been requested for a period of time in the past
Where t (f) is the set of base stations that have made requests for content f.Is a base station mjInterest in f of contentThe degree, here an element of the content popularity matrix P.
The specific caching mode in the step (3) is as follows:
(31) the content popularity (intra-class popularity) in each class is analyzed first, that is, the requested content of all the users served by the base station in the class is counted. The content is ordered by the number of accesses from high to low.
(32) Buffer capacity of each base station m is Sm. Eta is the cache capacity S occupied by the content cached through the intra-class popularitymPercentage of (c). Firstly, caching contents from high to low according to the intra-class traffic flow, and before caching the contents, checking whether the total size of the cached contents exceeds eta × S or notm. If yes, abandoning the cache.
(33) And for the residual buffer capacity of the base station, performing buffer by a step (22) based on base station cooperative filtering in the class, and performing buffer on the content from high to low according to p (m, f) until the total amount of the buffer content is larger than the buffer capacity.
While the invention has been shown and described with respect to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the scope of the invention as defined in the following claims.
Claims (1)
1. A caching method based on base station clustering is characterized by comprising the following steps:
(1) firstly, the problem of base station clustering is considered; counting the request frequency of each base station service user about the content in the past period, regarding each base station, taking the request frequency of the service user to each content as the characteristic of the base station, clustering the base stations by adopting k-means clustering, wherein the users served by the base stations of each class have similar interests and hobbies, namely the requested content has a plurality of similar places; the specific base station clustering comprises the following steps:
(11) according to historical request information of a period of time in the past, a content popularity matrix is obtained through analysis of data by a core networkWherein each element p in the matrixm,fRepresenting the frequency of requests for content f from users served by the base station m, the frequency of requests for content being characteristic of the base station, each row P of the matrix PmA vector representing an F dimension, representing a feature vector of a base station;
(12) randomly selecting k base stations as initial central points of base station clusters, and expressing feature vectors of the k base stations as initial central points Here, theSuperscript (1) denotes the first round, i.e. the initial value, and the subscript denotes the ith center point;
(13) from the center point of each class, to minimize the intra-class sum of squares, it is determined to which class the base station belongs as follows:here, theIndicating that the t-th round belongs to the base station set of the i-th class;
(15) repeating (13) and (14) to ciIs less than a given threshold, finally obtaining k class, H1,...,HkEach base station belongs to one of the classes;
(2) the Top N collaborative filtering recommendation system based on the base station predicts the content which is not requested by the user and covered by the base station by utilizing the similarity between the base stations; the cooperative filtering based on the base station specifically comprises the following steps:
(21) calculating similarity between base stations within classThe similarity between the intra-class base stations is calculated by the following similarity formula:
base station miAnd base station mjBelong to the same class, T (m)i) And T (m)j) Respectively represent base stations miAnd base station mjThe served users access a collection of content; t (f) represents a set of base stations that have accessed the content f;
(22) from (21), m can be obtainediSet S (m) of closest base stationsiG), then base station miThe served user has an interest level in the content f that has never been requested for a period of time in the past
Where t (f) is the set of base stations that have made requests for content f,is a base station mjThe degree of interest in f of the content, here the elements of the content popularity matrix P;
(3) giving each class in the first step, performing targeted caching on the base station by combining the collaborative filtering on each class according to the distribution of the content popularity in the class, and determining the content cached by each base station; the specific caching mode comprises the following steps:
(31) firstly, analyzing the content popularity in each class, namely counting the request contents of all base station service users in the class, and sequencing the request contents according to the access times of the contents from high to low;
(32) buffer capacity of each base station m is Sm(ii) a Eta is the cache capacity S occupied by the content cached through the intra-class popularitymThe content is firstly cached to the base station from high to low according to the intra-class traffic, and before caching the content, whether the total size of the cached content exceeds eta S is checkedmIf yes, giving up the cache;
(33) and for the residual buffer capacity of the base station, performing buffer by a step (22) based on base station cooperative filtering in the class, and performing buffer on the content from high to low according to p (m, f) until the total amount of the buffer content is larger than the buffer capacity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710704882.3A CN107592656B (en) | 2017-08-17 | 2017-08-17 | Caching method based on base station clustering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710704882.3A CN107592656B (en) | 2017-08-17 | 2017-08-17 | Caching method based on base station clustering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107592656A CN107592656A (en) | 2018-01-16 |
CN107592656B true CN107592656B (en) | 2020-12-11 |
Family
ID=61043151
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710704882.3A Active CN107592656B (en) | 2017-08-17 | 2017-08-17 | Caching method based on base station clustering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107592656B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108667653B (en) * | 2018-04-17 | 2020-12-11 | 东南大学 | Cluster-based cache configuration method and device in ultra-dense network |
CN108600365B (en) * | 2018-04-20 | 2020-05-22 | 西安交通大学 | Wireless heterogeneous network caching method based on sequencing learning |
CN108549719A (en) * | 2018-04-23 | 2018-09-18 | 西安交通大学 | A kind of adaptive cache method based on cluster in mobile edge calculations network |
CN108738048B (en) * | 2018-04-25 | 2021-02-26 | 杭州电子科技大学 | Active storage method of maximized fairness base station based on genetic algorithm |
CN108833352B (en) * | 2018-05-17 | 2020-08-11 | 北京邮电大学 | Caching method and system |
CN108990111B (en) * | 2018-06-13 | 2021-06-11 | 东南大学 | Base station caching method under condition that content popularity changes along with time |
CN109639844B (en) * | 2019-02-26 | 2020-06-05 | 北京中投视讯文化传媒股份有限公司 | Base station and content caching method based on local popularity |
CN112995979B (en) * | 2021-03-04 | 2022-01-25 | 中国科学院计算技术研究所 | Wireless network cache recommendation method for QoE (quality of experience) requirements of user |
CN113709816B (en) * | 2021-06-04 | 2024-03-22 | 武汉大学 | Base station collaborative caching method based on multi-feature user groups |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6651141B2 (en) * | 2000-12-29 | 2003-11-18 | Intel Corporation | System and method for populating cache servers with popular media contents |
CN103428267A (en) * | 2013-07-03 | 2013-12-04 | 北京邮电大学 | Intelligent cache system and method for same to distinguish users' preference correlation |
CN104955077A (en) * | 2015-05-15 | 2015-09-30 | 北京理工大学 | Heterogeneous network cell clustering method and device based on user experience speed |
CN106453495A (en) * | 2016-08-31 | 2017-02-22 | 北京邮电大学 | Information centric networking caching method based on content popularity prediction |
CN106714202A (en) * | 2015-11-16 | 2017-05-24 | 中国移动通信集团公司 | Network capacity optimization method and device |
-
2017
- 2017-08-17 CN CN201710704882.3A patent/CN107592656B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6651141B2 (en) * | 2000-12-29 | 2003-11-18 | Intel Corporation | System and method for populating cache servers with popular media contents |
CN103428267A (en) * | 2013-07-03 | 2013-12-04 | 北京邮电大学 | Intelligent cache system and method for same to distinguish users' preference correlation |
CN104955077A (en) * | 2015-05-15 | 2015-09-30 | 北京理工大学 | Heterogeneous network cell clustering method and device based on user experience speed |
CN106714202A (en) * | 2015-11-16 | 2017-05-24 | 中国移动通信集团公司 | Network capacity optimization method and device |
CN106453495A (en) * | 2016-08-31 | 2017-02-22 | 北京邮电大学 | Information centric networking caching method based on content popularity prediction |
Non-Patent Citations (1)
Title |
---|
内容流行度分布动态性对基站端缓存性能的影响;戚凯强等;《信号处理》;20170331;第33卷(第3期);参见第1-4部分 * |
Also Published As
Publication number | Publication date |
---|---|
CN107592656A (en) | 2018-01-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107592656B (en) | Caching method based on base station clustering | |
CN108667653B (en) | Cluster-based cache configuration method and device in ultra-dense network | |
CN110213627B (en) | Streaming media cache allocation method based on multi-cell user mobility | |
CN107909108A (en) | Edge cache system and method based on content popularit prediction | |
CN107968835B (en) | Wireless heterogeneous network video cache facility deployment method based on coding | |
CN108549719A (en) | A kind of adaptive cache method based on cluster in mobile edge calculations network | |
CN105812834B (en) | Video recommendations server, recommended method and pre-cache method based on clustering information | |
CN101287280A (en) | Network selecting method and device in heterogeneous wireless network | |
CN107277159B (en) | Ultra-dense network small station caching method based on machine learning | |
Jiang et al. | A novel caching policy with content popularity prediction and user preference learning in fog-RAN | |
CN108521640B (en) | Content distribution method in cellular network | |
Khan et al. | On the application of agglomerative hierarchical clustering for cache-assisted D2D networks | |
Bock et al. | A 2-step approach to improve data-driven parking availability predictions | |
CN108600365B (en) | Wireless heterogeneous network caching method based on sequencing learning | |
CN113472420B (en) | Satellite network cache placement method based on regional user interest perception | |
CN108810139A (en) | A kind of wireless caching method based on Monte Carlo tree search auxiliary | |
CN113271631A (en) | Novel content cache deployment scheme based on user request possibility and space-time characteristics | |
Wu et al. | Semigradient-based cooperative caching algorithm for mobile social networks | |
CN112862060A (en) | Content caching method based on deep learning | |
CN115361710A (en) | Content placement method in edge cache | |
Chakraborty et al. | R2-d2d: A novel deep learning based content-caching framework for d2d networks | |
CN113037872A (en) | Caching and prefetching method based on Internet of things multi-level edge nodes | |
CN111372096B (en) | D2D-assisted video quality adaptive caching method and device | |
Mishra et al. | Efficient proactive caching in storage constrained 5g small cells | |
CN112584439A (en) | Caching method in edge calculation |
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 |