CN116709422A - MEC task unloading method based on knowledge graph and matching theory - Google Patents

MEC task unloading method based on knowledge graph and matching theory Download PDF

Info

Publication number
CN116709422A
CN116709422A CN202310721151.5A CN202310721151A CN116709422A CN 116709422 A CN116709422 A CN 116709422A CN 202310721151 A CN202310721151 A CN 202310721151A CN 116709422 A CN116709422 A CN 116709422A
Authority
CN
China
Prior art keywords
user
server
mec
matching
knowledge graph
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310721151.5A
Other languages
Chinese (zh)
Inventor
刘伟
牛斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN202310721151.5A priority Critical patent/CN116709422A/en
Publication of CN116709422A publication Critical patent/CN116709422A/en
Pending legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/0231—Traffic management, e.g. flow control or congestion control based on communication conditions
    • H04W28/0236—Traffic management, e.g. flow control or congestion control based on communication conditions radio quality, e.g. interference, losses or delay
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/0268—Traffic management, e.g. flow control or congestion control using specific QoS parameters for wireless networks, e.g. QoS class identifier [QCI] or guaranteed bit rate [GBR]
    • 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
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an MEC task unloading method based on a knowledge graph and a matching theory, which mainly solves the problem of MEC task unloading under the condition that service quality data feedback of a user is limited to a certain extent due to factors such as environmental change and unstable network state. The implementation steps are as follows: constructing an MEC knowledge graph; vectorizing the MEC knowledge map through a TransE vector representation algorithm; calculating the similarity between users by using the vectorized MEC knowledge graph; predicting the service quality of each user; constructing a user preference list; constructing a server preference list; user matching selection; server matching selection; and judging whether the matching of the user and the server is finished, if so, realizing the task unloading of the user, and if not, continuing to execute the matching selection of the user.

Description

MEC task unloading method based on knowledge graph and matching theory
Technical Field
The invention belongs to the technical field of communication, and further relates to a MEC task unloading method based on a knowledge graph and a matching theory in the field of wireless communication. The invention can be applied to intelligent control of the communication network.
Background
At present, with the continuous popularization of 5G communication technology, a plurality of emerging application scenes such as automatic driving, digital twinning, industrial Internet of things and the like are continuously emerging. Mobile edge computing MEC (Mobile Edge Computing) is a network architecture that provides the services and computing functions required by the user on the wireless side, the user offloading tasks to servers in the MEC for task computation, where the primary performance indicator is user task latency. The main factors affecting MEC task offloading performance are two: one is that the scale and deployment location of resources in the MEC varies and is not evenly distributed across the network, and that multiple users in another MEC compete for server resources in the network. In a mobile edge computing network, user quality of service data feedback is limited due to factors such as environmental and network state instability. The traditional task offloading algorithm generally needs to acquire real-time information of user service quality under the current network, and then make task offloading decisions. The task offloading problem of a mobile edge computing network can be seen as a bilateral matching problem between a set of users and a set of servers. There are three categories of matching theory, one-to-one matching, many-to-one matching, and many-to-many matching. At present, most algorithms based on a matching theory in the problem of task offloading of a mobile edge computing network are of a many-to-one matching type, namely, a server can process a plurality of user tasks at the same time, and a user can only select one server at a time to offload tasks. The existing task offloading algorithm generally needs to acquire real-time information of user service quality under the current network first, and then make task offloading decisions. However, in an actual mobile edge computing network, the feedback of the service quality data of the user is limited due to factors such as environmental changes and unstable network states.
Chittaranjan Swain and Manmath Narayan Sahoo in its published paper "Spato: A student project allocation based task offloading in iot-fog systems" (ICC 2021-IEEE International Conference on Communications, 2021) propose a task offloading method based on student item matching theory. According to the method, the preference list of the user and the server is built through the analytic hierarchy process, and then the MEC task unloading algorithm is obtained through the bilateral matching process between the user and the server, so that the task time delay is effectively reduced. Although the method reduces the task time delay, the method still has the defects that firstly, the method needs to acquire the real-time information of the service quality of the user under the current network, and then makes a task unloading decision. However, in an actual mobile edge computing network, the feedback of the service quality data of the user is limited due to factors such as environmental changes and unstable network states. In the case of such conditions, the performance of the process may also be limited.
Disclosure of Invention
The invention aims to provide a MEC task unloading method based on a knowledge graph and a matching theory, aiming at overcoming the defects of the prior art, and solving the problem of MEC task unloading under the condition that the service quality data feedback of a user is limited to a certain extent due to factors such as environmental change and unstable network state.
The method comprises the steps of constructing a knowledge graph according to MEC historical data, vectorizing the knowledge graph through an existing knowledge graph vector expression algorithm, calculating user similarity through user vectors, predicting the target user service quality through the historical service quality of similar users, respectively constructing preference lists of users and servers according to the obtained user service quality data, and obtaining a task unloading scheme of the users through bilateral matching of a user set and a server set.
The method comprises the following specific steps:
step 1, constructing an MEC knowledge graph:
respectively constructing a user domain knowledge graph and a server domain knowledge graph according to historical data of MEC by using a mobile edge computing network; the user domain knowledge graph and the server domain knowledge graph form an MEC knowledge graph;
step 2, vectorizing the MEC knowledge graph through a TransE vector representation algorithm;
step 3, calculating the similarity between users by using the vectorized MEC knowledge graph;
step 4, predicting the service quality of each user to obtain a user service quality matrix;
step 5, constructing a user preference list:
sequencing the predicted service quality of the corresponding users on different servers in the predicted user service quality matrix according to the sequence from high to low, and forming a user u by M servers corresponding to the sequenced elements n Is (are) are (are) selected for use in the selection of a preferred list of(s)Wherein M > 0, u n Representing the first of MECn users;
step 6, constructing a server preference list:
sequencing the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix according to the sequence from high to low, and forming N users corresponding to the sequenced elements in the server ES m Is set of preference lists of (a)Wherein N > 0, ES m Representing an mth server in the MEC;
step 7, user matching selection:
traversing the user set U, if the current user U n Is not matched with any server, and is in the user preference listThe server with the highest priority is selected +.>Associating it with user u n The corresponding location element in the matching matrix Ω is set to 1 and the server is moved out of the current user u n Is a preferred list of (2); if the current user is matched, the user can only load the task to at most one server, so that the user does not perform matching selection any more;
step 8, server matching selection:
traversing the server set ES, if the server ES is currently selected m The number of users of (a) has exceeded the load gamma of the server m According to the preference list of the serverThe user with the lowest priority among the currently matched users is +.>Removing from the matching matrix omega and setting the corresponding location element to 0 until the load of the server is satisfied; if the service is currently selectedThe number of users of the server does not reach the server load upper limit gamma m No operation is performed, wherein γ m ＞0；
Step 9, judging whether all users are matched or not, or whether a user preference list of incomplete matching is empty, if yes, executing step 10, otherwise, executing step 7;
and step 10, completing the many-to-one bilateral matching between the user set and the server set, and realizing the task unloading of the user.
Compared with the prior art, the invention has the following advantages:
firstly, the bilateral matching task unloading algorithm based on the knowledge spectrum, which is provided by the invention, constructs the knowledge spectrum according to the mobile edge computing network data, and further vectorizes the triplet of the knowledge spectrum through a TransE vector representation algorithm. And then, calculating the user similarity by using the user vector, and predicting the user service quality data. On the basis, preference lists of users and servers are respectively constructed, and bilateral matching of the user sets and the server sets is carried out.
Second, because the data used in the invention are all historical service quality data, the limitation that the service quality data feedback of the user is limited due to factors such as environmental change and unstable network state in the actual mobile edge computing network is overcome, and the task unloading can be performed under the condition of limited conditions.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of a user domain knowledge graph in the present invention;
fig. 3 is a schematic diagram of a knowledge graph of a server domain in the present invention.
Detailed Description
The invention is further described below with reference to the drawings and examples.
The implementation steps of the embodiment of the present invention will be further described with reference to fig. 1.
And step 1, constructing an MEC knowledge graph.
Using a mobile edge computing network, according to historical data of MEC, wherein N pieces of user equipment U= { U 1 ,u 2 ,...,u n ,...,u N M servers es= { ES 1 ,ES 2 ,...,ES m ,...,ES M And (3) jointly forming. Wherein M > 5,N > 10, u n Represents the nth user, ES m Representing an mth server, and respectively constructing a user domain knowledge graph and a server domain knowledge graph; and forming the MEC knowledge graph by the user domain knowledge graph and the server domain knowledge graph.
The user domain knowledge graph comprises an attribute layer and a data layer, as shown in fig. 2.
The attribute layer is composed of category attributes of network attributes, task attributes, location attributes and equipment attributes, for example, the network attributes of the circle user 1 in fig. 2 correspond to the circle IP indicated by the network attribute arrow in fig. 2, etc., and the data layer is a numerical value corresponding to each attribute, for example, the value of the circle IP in fig. 2 corresponds to the circle 137.189.97.17 indicated by the corresponding arrow in fig. 2.
The server domain knowledge graph is divided into an attribute layer and a data layer, as shown in fig. 3.
The attribute layer is formed by various attributes such as network attribute, resource attribute, location attribute, etc., for example, the network attribute of the circle server 1 in fig. 3 corresponds to the circle IP indicated by the network attribute arrow in fig. 3, etc., and the data layer is a numerical value corresponding to each attribute, for example, the value of the circle IP in fig. 3 corresponds to the circle 137.189.98.31 indicated by the corresponding arrow in fig. 3.
Step 2, vectorizing the MEC knowledge graph through a TransE vector expression algorithm to obtain k-dimensional vector expression of the user, wherein k is valued according to the vector expression precision, and k is more than 10, such as user u n Is expressed as { v } n,1 ,v n,2 ,...,v n,k }。
And 3, calculating the similarity between users by using the vectorized MEC knowledge graph.
The similarity between the users is obtained by the following formula:
wherein sim (u) n1 ,u n2 ) Representing user u n1 With user u n2 Similarity between, similarity sim (u n1 ,u n2 )∈[0,1]K is the vector dimension after knowledge graph vectorization, user u n1 Is { v } n1,1 ,v n1,2 ,...,v n1,k User u n2 Is { v } n2,1 ,v n2,2 ,...,v n2,k }，v n1,i For user u n1 I-th dimension vector, v n2,i For user u n2 Obtaining a user similarity matrix Simu, and the similarity between all users corresponding to the matrix is shown in the following table.
And 4, predicting the service quality of each user to obtain a user service quality matrix.
For target user u n At the server ES m Quality of service of (c)Predicting, according to the user similarity matrix Simu, the method comprises the steps of determining a server ES m Of the users of quality of service, the selection and target user u n The alpha users with highest similarity form the target user u n Is set of similar users S (u) n )＝{u n1 ,u n2 ,...,u nα And the value range of alpha is alpha > 5. Then, the server ES is accessed by the similar users in the collection m Historical quality of service calculations of (a).
The prediction of the unknown quality of service is derived from the following equation:
wherein ,representing user u n At the server ES m Predicted quality of service for task calculation, +.>For user u n Average quality of service, q, across different servers i,m Representing user u i At the server ES m Historical quality of service data for task calculation, +.>For user u i Is an average quality of service on different servers, S (u n ) For user u n Is a set of similar users.
Step 5, constructing a user preference list:
sequencing the predicted service quality of the corresponding users on different servers in the predicted user service quality matrix according to the sequence from high to low, and forming a user u by M servers corresponding to the sequenced elements n Is (are) are (are) selected for use in the selection of a preferred list of(s)Wherein M > 0, u n Representing the nth user in the MEC.
And 6, constructing a server preference list.
Sequencing the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix according to the sequence from high to low, and forming N users corresponding to the sequenced elements in the server ES m Is set of preference lists of (a)Wherein N > 0, ES m Represents the mth in MECAnd a server.
And 7, matching and selecting by a user.
Traversing the user set U, if the current user U n Is not matched with any server, and is in the user preference listThe server with the highest priority is selected +.>Associating it with user u n The corresponding location element in the matching matrix Ω is set to 1 and the server is moved out of the current user u n Is a preferred list of (2); if the current user has matched, the user does not make a matching selection anymore, since the user can only offload tasks to at most one server.
The matching matrix Ω is a result of recording the matching between the user and the server, and is initialized to be an all-zero matrix, Ω, representing a real matrix, N is the number of rows of the real matrix, the value of which is equal to the total number of users, M is the number of columns of the real matrix, and the value of which is equal to the total number of servers.
And 8, matching and selecting the server.
Traversing the server set ES, if the server ES is currently selected m The number of users of (a) has exceeded the load gamma of the server m According to the preference list of the serverThe user with the lowest priority among the currently matched users is +.>Removed from the matching matrix Ω and the corresponding location element is set to 0 until the load of the server is satisfied. If when it isThe number of users previously selecting the server does not reach the upper server load limit gamma m No operation is performed, wherein γ m ＞0。
And 9, judging whether all users are matched and complete matching or whether a user preference list of incomplete matching is empty, if so, executing the step 10, otherwise, executing the step 7.
And step 10, completing the many-to-one bilateral matching between the user set and the server set, and realizing the task unloading of the user.

Claims (6)

1. The MEC task unloading method based on the knowledge graph and the matching theory is characterized by comprising the steps of constructing an MEC knowledge graph, respectively constructing preference lists of users and servers, and completing task unloading of the users through bilateral matching of a user set and a server set; the task unloading method comprises the following steps:
step 1, constructing an MEC knowledge graph:
respectively constructing a user domain knowledge graph and a server domain knowledge graph according to historical data of MEC by using a mobile edge computing network; the user domain knowledge graph and the server domain knowledge graph form an MEC knowledge graph;
step 2, vectorizing the MEC knowledge graph through a TransE vector representation algorithm;
step 3, calculating the similarity between users by using the vectorized MEC knowledge graph;
step 4, predicting the service quality of each user to obtain a user service quality matrix;
step 5, constructing a user preference list:
sequencing the predicted service quality of the corresponding users on different servers in the predicted user service quality matrix according to the sequence from high to low, and forming a user u by M servers corresponding to the sequenced elements n Is (are) are (are) selected for use in the selection of a preferred list of(s)Wherein M > 0, u n Representing the nth user in the MEC;
step 6, constructing a server preference list:
sequencing the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix according to the sequence from high to low, and forming N users corresponding to the sequenced elements in the server ES m Is set of preference lists of (a)Wherein N > 0, ES m Representing an mth server in the MEC;
step 7, user matching selection:
traversing the user set U, if the current user U n Is not matched with any server, and is in the user preference listThe server with the highest priority is selected +.>Associating it with user u n The corresponding location element in the matching matrix Ω is set to 1 and the server is moved out of the current user u n Is a preferred list of (2); if the current user is matched, the user can only load the task to at most one server, so that the user does not perform matching selection any more;
step 8, server matching selection:
traversing the server set ES, if the server ES is currently selected m The number of users of (a) has exceeded the load gamma of the server m According to the preference list of the serverThe user with the lowest priority among the currently matched users is +.>Remove from the matching matrix Ω and set the corresponding location element to 0 until the server is satisfiedLoading; if the number of users currently selecting the server does not reach the upper limit gamma of the server load m No operation is performed, wherein γ m ＞0；
Step 9, judging whether all users are matched or not, or whether a user preference list of incomplete matching is empty, if yes, executing step 10, otherwise, executing step 7;
and step 10, completing the many-to-one bilateral matching between the user set and the server set, and realizing the task unloading of the user.
2. The MEC task offloading method according to claim 1, wherein the user domain knowledge graph in step 1 includes an attribute layer and a data layer, the attribute layer is composed of a network attribute, a location attribute, a user attribute, and a class attribute of a task attribute, and the data layer is a numerical value corresponding to each attribute.
3. The MEC task offloading method based on knowledge graph and matching theory of claim 1, wherein the server domain knowledge graph in step 1 is divided into an attribute layer and a data layer; the attribute layer is composed of various attributes such as network attributes, position attributes, server attributes and the like; the data layer is composed of { computing resources, computing resource data, 8-core CPU }, { memory resources, memory resource data, 200G } triples.
4. The MEC task offloading method of claim 1, wherein the similarity between the users in step 3 is obtained by:
wherein sim (u) n1 ,u n2 ) Representing user u n1 With user u n2 Similarity between the two, k is the vector dimension after the vectorization of the knowledge graphDegree, user u n1 Is { v } n1,1 ,v n1,2 ,...,v n1,k User u n2 Is { v } n2,1 ,v n2,2 ,...,v n2,k }。
5. The method for MEC task offloading based on knowledge-graph and matching theory of claim 4, wherein the predicting of the unknown quality of service in step 4 is obtained by:
wherein ,representing user u n At the server ES m Predicted quality of service for task calculation, +.>For user u n Average quality of service, q, across different servers i,m Representing user u i At the server ES m Historical quality of service data for performing the task calculations,for user u i Is an average quality of service on different servers, S (u n ) For user u n Is a set of similar users.
6. The method for offloading MEC tasks based on knowledge graph and matching theory according to claim 1, wherein in step 7, the matching matrix Ω is a matching result of the recording user and the server, the matching matrix Ω is initialized to be an all-zero matrix, Ω, representing a real matrix, N is the number of rows of the real matrix, the value of which is equal to the total number of users, M is the number of columns of the real matrix, and the value of which is equal to the total number of servers.
CN202310721151.5A 2023-06-16 2023-06-16 MEC task unloading method based on knowledge graph and matching theory Pending CN116709422A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310721151.5A CN116709422A (en) 2023-06-16 2023-06-16 MEC task unloading method based on knowledge graph and matching theory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310721151.5A CN116709422A (en) 2023-06-16 2023-06-16 MEC task unloading method based on knowledge graph and matching theory

Publications (1)

Publication Number Publication Date
CN116709422A true CN116709422A (en) 2023-09-05

Family

ID=87844802

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310721151.5A Pending CN116709422A (en) 2023-06-16 2023-06-16 MEC task unloading method based on knowledge graph and matching theory

Country Status (1)

Country Link
CN (1) CN116709422A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118034920A (en) * 2024-01-29 2024-05-14 广东技术师范大学 Network computing resource collaborative scheduling method integrating user intention and knowledge graph

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118034920A (en) * 2024-01-29 2024-05-14 广东技术师范大学 Network computing resource collaborative scheduling method integrating user intention and knowledge graph

Similar Documents

Publication Publication Date Title
CN111242282B (en) Deep learning model training acceleration method based on end edge cloud cooperation
CN106446959B (en) A kind of cloud computing resources dynamic matching method and device
CN112181666A (en) Method, system, equipment and readable storage medium for equipment evaluation and federal learning importance aggregation based on edge intelligence
CN111079780B (en) Training method for space diagram convolution network, electronic equipment and storage medium
CN110458084B (en) Face age estimation method based on inverted residual error network
CN111104831B (en) Visual tracking method, device, computer equipment and medium
CN116709422A (en) MEC task unloading method based on knowledge graph and matching theory
EP4339843A1 (en) Neural network optimization method and apparatus
US20240233358A9 (en) Image classification method, model training method, device, storage medium, and computer program
CN109325530B (en) Image classification method, storage device and processing device
CN115952832A (en) Adaptive model quantization method and apparatus, storage medium, and electronic apparatus
CN114724021B (en) Data identification method and device, storage medium and electronic device
CN114511042A (en) Model training method and device, storage medium and electronic device
CN113504949B (en) Task unloading and parameter optimization method and system for MAR client in edge calculation
CN116541712A (en) Federal modeling method and system based on non-independent co-distributed data
Singhal et al. Resource-aware Deployment of Dynamic DNNs over Multi-tiered Interconnected Systems
CN115955685A (en) Multi-agent cooperative routing method, equipment and computer storage medium
CN113157453B (en) Task complexity-based high-energy-efficiency target detection task dynamic scheduling method
Onishi et al. Memory efficient training using lookup-table-based quantization for neural network
CN114363671A (en) Multimedia resource pushing method, model training method, device and storage medium
CN114547473A (en) Crowdsourcing task data recommendation method based on decision model and genetic matrix decomposition method
CN113537492A (en) Model training and data processing method, device, equipment, medium and product
CN113065641A (en) Neural network model training method and device, electronic equipment and storage medium
CN112685176A (en) Resource-constrained edge computing method for improving DDNN (distributed neural network)
CN113033653A (en) Edge-cloud collaborative deep neural network model training method

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