MEC Task Unloading Method Based on Knowledge Graph and Matching Theory

Abstract

This invention discloses a MEC task offloading method based on knowledge graph and matching theory, mainly addressing the MEC task offloading problem when user service quality data feedback is limited due to environmental changes and network instability. The implementation steps are: constructing an MEC knowledge graph; vectorizing the MEC knowledge graph using the TransE vector representation algorithm; calculating the similarity between users using the vectorized MEC knowledge graph; predicting the service quality for each user; constructing a user preference list; constructing a server preference list; user matching selection; server matching selection; determining whether user and server matching is complete; if so, implementing user task offloading; otherwise, continuing with user matching selection.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and more specifically relates to a MEC task offloading method based on knowledge graph and matching theory in the field of wireless communication. This invention can be applied to the intelligent management and control of communication networks. Background Technology

[0002] Currently, with the continuous popularization of 5G communication technology, many emerging application scenarios such as autonomous driving, digital twins, and industrial IoT are constantly emerging. Mobile edge computing (MEC) is a network architecture that provides users with the services and computing functions they need on the wireless side. Users offload tasks to servers in the MEC for task computation, and the main performance indicator is user task latency. There are two main factors affecting MEC task offloading performance: one is the difference in the scale and deployment location of resources in the MEC, which is not uniformly distributed in the network; the other is the competition among multiple users in the MEC for server resources in the network. In mobile edge computing networks, due to factors such as unstable environment and network status, the feedback of user service quality data is limited to a certain extent. Traditional task offloading algorithms usually need to obtain real-time information on user service quality under the current network before making task offloading decisions. The task offloading problem in mobile edge computing networks can be regarded as a two-sided matching problem between the user set and the server set. Matching theory has three classifications: one-to-one matching, many-to-one matching, and many-to-many matching. Currently, most algorithms based on matching theory in the task offloading problem of mobile edge computing networks are of the many-to-one matching type, meaning that a server can handle multiple user tasks simultaneously, but a user can only choose one server to offload their task at a time. Existing task offloading algorithms typically require obtaining real-time information on the user's quality of service (QoS) under the current network conditions before making a task offloading decision. However, in real-world mobile edge computing networks, user QoS data feedback is limited due to factors such as environmental changes and network instability.

[0003] In their paper "Spato: Astudent project allocation based task offloading in iot-fog systems" (ICC2021-IEEE International Conference on Communications, 2021), Chittaranjan Swain and Manmath Narayan Sahoo proposed a task offloading method based on student project matching theory. This method constructs user and server preference lists using the analytic hierarchy process (AHP), and then derives an MEC task offloading algorithm through a bilateral matching process between users and servers, effectively reducing task latency. While this method reduces latency, it still has limitations. First, it requires real-time information on user service quality under the current network conditions before making task offloading decisions. However, in real-world mobile edge computing networks, user service quality data feedback is limited due to environmental changes and network instability. Under these constraints, the performance of this method is also limited. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the existing technology by providing a MEC task offloading method based on knowledge graphs and matching theory. This method aims to solve the MEC task offloading problem when user service quality data feedback is limited due to factors such as environmental changes and unstable network conditions.

[0005] The idea behind this invention is to construct a knowledge graph based on MEC historical data, then vectorize the knowledge graph using existing knowledge graph vector representation algorithms, calculate user similarity using user vectors, predict the service quality of the target user based on the historical service quality of similar users, construct user and server preference lists based on the obtained user service quality data, and obtain the user's task unloading scheme through bilateral matching of the user set and the server set.

[0006] The specific steps for implementing this invention are as follows:

[0007] Step 1, Construct MEC Knowledge Graph:

[0008] Using mobile edge computing networks, user domain knowledge graphs and server domain knowledge graphs are constructed based on historical MEC data; the user domain knowledge graph and the server domain knowledge graph are then combined to form the MEC knowledge graph.

[0009] Step 2: Vectorize the MEC knowledge graph using the TransE vector representation algorithm;

[0010] Step 3: Calculate the similarity between users using the vectorized MEC knowledge graph;

[0011] Step 4: Predict the service quality for each user to obtain the user service quality matrix;

[0012] Step 5, build a user preference list:

[0013] Sort the predicted service quality of the corresponding users in the predicted user service quality matrix on different servers in descending order, and then sort the elements corresponding to the predicted service quality. A user group consists of servers Preference list ,in, , Indicates the first in MEC One user;

[0014] Step 6, Build the server preference list:

[0015] Sort the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix from high to low, and then sort the elements corresponding to the sorted elements. A group of users on the server Preference list collection ,in, , Indicates the first in MEC One server;

[0016] Step 7, User Matching Selection:

[0017] Traverse the user set If the current user No match was found in the user preference list. Select the server with the highest priority to connect with users In the matching matrix Set the corresponding element in the middle to 1, and remove the server from the current user's list. The user's preference list; if the current user has already matched, since the user can only unload tasks to at most one server, the user will not make any further matching selections;

[0018] Step 8, Server Matching Selection:

[0019] Traverse the server collection If this server is selected at the moment The number of users has exceeded the server's capacity. Then, based on the server's preference list... The user with the lowest priority among the currently matched users. From the matching matrix Remove elements from the list and set the corresponding element to 0 until the server load is met; if the number of users currently selecting the server has not reached the server load limit. No operation is performed, among which, ;

[0020] Step 9: Determine whether all users have been matched or whether the list of users whose preferences have not been matched is empty. If yes, proceed to step 10; otherwise, proceed to step 7.

[0021] Step 10: Complete the many-to-one bilateral matching between the user set and the server set to enable the user's task unloading.

[0022] Compared with existing technologies, the present invention has the following advantages:

[0023] First, the knowledge graph-based bilateral matching task offloading algorithm proposed in this invention constructs a knowledge graph based on mobile edge computing network data, and then vectorizes the knowledge graph triples using the TransE vector representation algorithm. Next, user similarity is calculated using user vectors to predict user service quality data. Based on this, preference lists for users and servers are constructed separately, and then bilateral matching of the user set and server set is performed. Compared to traditional methods that directly use user service quality information to construct preference lists, the method proposed in this invention considers multiple user attributes, resulting in a superior preference list.

[0024] Secondly, since the data used in this invention are all historical service quality data, it overcomes the limitation that user service quality data feedback is restricted in actual mobile edge computing networks due to factors such as environmental changes and unstable network conditions. Task offloading can also be performed under limited conditions. Attached Figure Description

[0025] Figure 1 This is a flowchart of the present invention;

[0026] Figure 2 This is a schematic diagram of the user domain knowledge graph in this invention;

[0027] Figure 3 This is a schematic diagram of the server domain knowledge graph in this invention. Detailed Implementation

[0028] The present invention will now be further described with reference to the accompanying drawings and embodiments.

[0029] Reference Figure 1The implementation steps of the embodiments of the present invention will be further described below.

[0030] Step 1: Construct the MEC knowledge graph.

[0031] Utilizing mobile edge computing networks, based on historical MEC data, where... One user terminal device as well as One server Together they constitute. Among them, , , Indicates the first One user, Indicates the first Each server constructs a user domain knowledge graph and a server domain knowledge graph respectively; the user domain knowledge graph and the server domain knowledge graph are combined to form an MEC knowledge graph.

[0032] The user domain knowledge graph includes an attribute layer and a data layer, such as Figure 2 As shown.

[0033] The attribute layer consists of category attributes for network attributes, task attributes, location attributes, and device attributes, for example... Figure 2 The network attribute of user 1 in the circle corresponds to Figure 2 The circled IP address indicated by the arrow in the network attribute section represents the data layer, which contains the numerical value corresponding to each attribute. Figure 2 The value of IP in the middle circle corresponds to Figure 2 The circle indicated by the corresponding arrow is 137.189.97.17.

[0034] The server domain knowledge graph is divided into an attribute layer and a data layer, such as Figure 3 As shown.

[0035] The attribute layer consists of various attributes such as network attributes, resource attributes, and location attributes, for example... Figure 3 The network attributes of circle server 1 in the image correspond to... Figure 3 The circled IP address indicated by the arrow in the network attribute section represents the data layer, which contains the numerical value corresponding to each attribute. Figure 3 The value of IP in the middle circle corresponds to Figure 3 The circle indicated by the corresponding arrow is 137.189.98.31.

[0036] Step 2: Vectorize the MEC knowledge graph using the TransE vector representation algorithm to obtain the user's... 3D vector representation, where the accuracy of vector representation depends on... To retrieve values, , such as users The vector representation is .

[0037] Step 3: Calculate the similarity between users using the vectorized MEC knowledge graph.

[0038] The similarity between users is obtained by the following formula:

[0039] ;

[0040] in, Indicates user With users Similarity between them , For the vector dimension after knowledge graph vectorization, the user The vector is ,user The vector is , For users The dimensional vector, For users The A dimensional vector is used to obtain the user similarity matrix. The similarity among all users corresponding to this matrix is ​​shown in the table below.

[0041]

[0042] Step 4: Predict the service quality for each user to obtain the user service quality matrix.

[0043] For target users On the server service quality Make predictions based on the user similarity matrix. In the known server Among users of service quality, select those who are the target users. Highest similarity The target users are composed of 10 users. similar user set ,in, The range of values ​​is Then, similar users in that set are analyzed on the server. Historical service quality calculation.

[0044] The prediction of unknown service quality is obtained by the following formula:

[0045] ;

[0046] in, Indicates user On the server Predictive service quality for task computation. For users The average service quality across different servers Indicates user On the server Historical service quality data for task calculation. For users The average service quality across different servers For users A set of similar users.

[0047] Step 5, build a user preference list:

[0048] Sort the predicted service quality of the corresponding users in the predicted user service quality matrix on different servers in descending order, and then sort the elements corresponding to the predicted service quality. A user group consists of servers Preference list ,in, , Indicates the first in MEC One user.

[0049] Step 6: Build a list of server preferences.

[0050] Sort the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix from high to low, and then sort the elements corresponding to the sorted elements. A group of users on the server Preference list collection ,in, , Indicates the first in MEC One server.

[0051] Step 7, User matching selection.

[0052] Traverse the user set If the current user No match was found in the user preference list. Select the server with the highest priority to connect with users In the matching matrix Set the corresponding element in the middle to 1, and remove the server from the current user's list. The user's preference list; if the current user has already matched, since the user can only unload tasks to at most one server, the user will not make any further matching selections.

[0053] The matching matrix To record the matching results between users and the server, a matching matrix is ​​used. Initialize as a matrix of all zeros. , Represents a real matrix. The row number of the real number matrix is ​​equal to the total number of users. The column number of the real number matrix is ​​equal to the total number of servers.

[0054] Step 8, server matching selection.

[0055] Traverse the server collection If this server is selected at the moment The number of users has exceeded the server's capacity. Then, based on the server's preference list... The user with the lowest priority among the currently matched users. From the matching matrix Remove elements from the list and set the corresponding element to 0, continuing until the server load is met. If the number of users currently selecting the server has not reached the server's load limit... No operation is performed, among which, .

[0056] Step 9: Determine whether all users have been matched or whether the list of user preferences that have not been matched is empty. If yes, proceed to step 10; otherwise, proceed to step 7.

[0057] Step 10: Complete the many-to-one bilateral matching between the user set and the server set to enable the user's task unloading.

Claims

1. A method for unloading MEC tasks based on knowledge graphs and matching theory, characterized in that, A MEC knowledge graph is constructed, and preference lists for users and servers are built separately. Through bilateral matching of the user set and the server set, task unloading for users is completed. The steps of this task unloading method include the following: Step 1, Construct MEC Knowledge Graph: Using mobile edge computing networks, user domain knowledge graphs and server domain knowledge graphs are constructed based on historical MEC data; the user domain knowledge graph and the server domain knowledge graph are then combined to form the MEC knowledge graph. Step 2: Vectorize the MEC knowledge graph using the TransE vector representation algorithm; Step 3: Calculate the similarity between users using the vectorized MEC knowledge graph, using the following formula: ； in, Indicates user With users Similarity between them This represents the vector dimension after the knowledge graph has been vectorized, for the user. The vector is ,user The vector is ; Step 4: Predict the service quality for each user to obtain the user service quality matrix, the formula of which is as follows: ； in, Indicates user On the server Predictive service quality for task computation. For users The average service quality across different servers Indicates user On the server Historical service quality data for task calculation. Indicates user The average service quality across different servers Indicates user A set of similar users; With target users Highest similarity One user, The range of values ​​is ; Step 5, build a user preference list: Sort the predicted service quality of the corresponding users in the predicted user service quality matrix on different servers in descending order, and then sort the elements corresponding to the predicted service quality. A user group consists of servers Preference list ,in, , Indicates the first in MEC One user; Step 6, Build the server preference list: Sort the predicted service quality of different users on the corresponding servers in the predicted user service quality matrix from high to low, and then sort the elements corresponding to the sorted elements. A group of users on the server Preference list collection ,in, , Indicates the first in MEC One server; Step 7, User Matching Selection: Traverse the user set If the current user No match was found in the user preference list. Select the server with the highest priority to connect with users In the matching matrix Set the corresponding element in the middle to 1, and remove the server from the current user's list. The user's preference list; if the current user has already matched, since the user can only unload tasks to at most one server, the user will not make any further matching selections; Step 8, Server Matching Selection: Traverse the server collection If this server is selected at the moment The number of users has exceeded the server's capacity. Then, based on the server's preference list... The user with the lowest priority among the currently matched users. From the matching matrix Remove elements from the list and set the corresponding element to 0 until the server load is met; if the number of users currently selecting the server has not reached the server load limit. No operation is performed, among which, ; Step 9: Determine whether the current match meets the termination condition. If yes, proceed to step 10; otherwise, proceed to step 7. The matching termination condition is that all users have completed the matching or the user preference list is empty. Step 10: Complete the many-to-one bilateral matching between the user set and the server set to enable the user's task unloading.

2. The MEC task offloading method based on knowledge graph and matching theory according to claim 1, characterized in that, The user domain knowledge graph mentioned in step 1 includes an attribute layer and a data layer. The attribute layer consists of category attributes such as network attributes, location attributes, user attributes, and task attributes, while the data layer consists of the numerical values ​​corresponding to each attribute.

3. The MEC task offloading method based on knowledge graph and matching theory according to claim 1, characterized in that, The server domain knowledge graph mentioned in step 1 is divided into an attribute layer and a data layer; the attribute layer consists of various attributes such as network attributes, location attributes, and server attributes; the data layer consists of triples of {computing resources, computing resource data, 8-core CPU} and {storage resources, storage resource data, 200G}.

4. The MEC task offloading method based on knowledge graph and matching theory according to claim 1, characterized in that, The matching matrix mentioned in step 7 To record the matching results between users and the server, a matching matrix is ​​used. Initialize as a matrix of all zeros. , Represents a real matrix. This represents the row number of the real number matrix, and its value is equal to the total number of users. This represents the number of columns in a real matrix, and its value is equal to the total number of servers.

Images