A method for pre-deploying edge services based on location prediction

Abstract

The present invention discloses an edge service pre-deployment method based on location prediction, which belongs to the field of service deployment based on mobile edge computing. The present invention uses the shape information and spatiotemporal information of user trajectories to judge the trajectory similarity, screens out a set of paths that meet the similarity criteria from the user's historical data, and establishes a transition probability matrix to predict the user's location at the next time point, which significantly increases the accuracy of user location prediction while ensuring the service time; at the same time, with the user as the center, a preference list between the user and the mobile edge computing MEC server is established, and bilateral stable matching between the user and the MEC server is achieved at the predicted location, and the service pre-deployment problem is modeled as a mean field game model, which reduces the service pre-deployment delay while increasing the user data transmission rate, thereby improving the user experience.

Description

Technical Field

[0001] The present invention relates to the field of service deployment based on mobile edge computing, and in particular to an edge service pre-deployment method based on location prediction. Background Art

[0002] In recent years, with the popularity of smart phones and the development of mobile Internet, people's demand for data services has been growing. At the same time, the widespread application of IoT technology has made a large number of devices need to access the network for data transmission and processing. This puts higher requirements on data processing capabilities, transmission speed and bandwidth, and requires a more efficient and fast computing and transmission method. In many application scenarios, such as autonomous driving, telemedicine, industrial control, etc., real-time data processing and low latency are key factors. The traditional cloud computing model is difficult to meet the needs of these applications due to problems such as data transmission delay and network congestion. Therefore, a technology that can perform real-time data processing at the edge of the network is needed. Mobile Edge Computing (MEC) is a technology that deploys computing resources, data storage and processing capabilities at the edge of the mobile network. It can provide mobile users with the services and cloud computing functions required by the nearest wireless access network. MEC can create a telecom-grade service environment with high performance, low latency and high bandwidth to meet the needs of real-time data processing and low latency. One of the core goals of MEC is to reduce the delay of data transmission and ensure the seamless operation of real-time applications.

[0003] At present, in the field of mobile edge computing, researchers have conducted in-depth and comprehensive explorations on multiple key issues. Among them, a highly anticipated research direction is to combine MEC technology to continuously promote and optimize the service pre-deployment problem for mobile users. This research direction aims to predict the future location and behavior of mobile users and deploy corresponding services on edge servers in advance to reduce service latency, improve user experience, and optimize resource allocation and system efficiency. Specifically, by predicting the user's location and deploying services on the edge server in the predicted area in advance, the edge device can collect user location information in real time and transmit the data to the cloud platform through the edge gateway for analysis and prediction. Once the user's location is predicted, the cloud platform can pre-deploy the required services to the edge server in the predicted area in advance, thereby ensuring that the service can be provided quickly when the user needs it. In addition, predicting the user's location can help the system predict future resource requirements more accurately, thereby optimizing resource allocation. However, in the field of MEC, a major pain point in the research of service pre-deployment technology based on user location prediction is how to efficiently use the analysis of user historical data and the learning of user mobility patterns to achieve fast and accurate path prediction. At the same time, designing a reasonable service pre-deployment strategy is the key to ensuring that deployment latency is minimized while meeting service requirements. This requires that the strategy can comprehensively consider multi-dimensional factors such as network conditions, resource availability, and user behavior, and achieve early deployment and rapid response of services through precise timing analysis and resource scheduling. In addition, optimizing resource utilization efficiency is also a research issue in service pre-deployment technology. Researchers need to explore how to dynamically adjust the number and configuration of service instances to adapt to the personalized needs of different users and different time periods, thereby bringing users a smoother and more efficient service experience.

[0004] In summary, using user historical data analysis and mobile pattern learning to predict paths, designing reasonable service pre-deployment strategies, and optimizing resource utilization efficiency are key issues that need to be urgently addressed in the current research on service pre-deployment technology in MEC. Solving these problems will help improve the response speed of mobile services, reduce latency, and provide users with a more stable and efficient service experience.

[0005] At present, some researchers have built prediction models to predict users' future service needs by collecting and analyzing the historical location information and access records of mobile users. And deploy corresponding service resources in MEC nodes in advance based on the prediction results. However, this processing solution will lead to two deficiencies: first, due to the variability and uncertainty of user behavior, the prediction model that only considers the spatiotemporal information of historical trajectories cannot accurately predict the user's location, which will lead to waste or insufficient service resources; second, this solution ignores the impact of the preference between users and MEC servers on the service pre-deployment strategy.

[0006] In addition, in response to the problem of single data in location prediction, some researchers have proposed a service prediction solution that integrates multi-source data. In addition to user historical data, data from other sources (such as social network data, weather data, etc.) are also integrated. These multi-source data can provide more information and context for the prediction model, thereby improving the accuracy and comprehensiveness of service predictions. However, this processing solution also has some shortcomings: First, the solution does not consider the complexity of data fusion. Fusion of multi-source data requires processing the heterogeneity and inconsistency of data, and MEC servers with weak computing power cannot withstand the complexity and difficulty of such data processing. Secondly, data from different sources may have quality differences and reliability issues, and appropriate data cleaning and verification measures need to be taken to ensure data quality, which will further increase the computing burden of MEC servers and increase edge service latency. Summary of the invention

[0007] The present invention proposes an edge service pre-deployment method based on location prediction, so as to maximize the user's data transmission rate and minimize the service pre-deployment delay under the premise of realizing the location prediction of mobile users and meeting service requirements.

[0008] The technical solution adopted by the present invention is:

[0009] A method for pre-deploying edge services based on location prediction, the method comprising the following steps:

[0010] Collect the user's historical trajectory and obtain the user's historical trajectory set for each user;

[0011] Filter out the user historical trajectories whose trajectory similarity with the user's current trajectory is greater than or equal to the similarity threshold from the user historical trajectory set of each user, and construct the trajectory transition probability matrix of the current user based on the filtered user historical trajectories;

[0012] Predicting the predicted position of each user based on the trajectory transition probability matrix of each user;

[0013] Based on the predicted location of each user, for each mobile edge computing server (MEC server) to be deployed, if the current user is within the communication range of the current mobile edge computing server, the mobile edge computing server is added to the candidate mobile edge computing server set of the current user;

[0014] Under the premise of satisfying the upper limit of edge computing servers that a single user can match in the same time slot and the upper limit of the number of users that a single edge computing server can serve, the server matching optimization goal is to maximize the total transmission efficiency of all current users, obtain the bilateral matching results of users and mobile edge computing servers in the set of candidate mobile edge computing servers, and obtain the matching mobile edge computing server for each user;

[0015] On the premise of satisfying the user's service request, the service pre-deployment optimization goal is to minimize the service pre-deployment delay, and pre-deploy the services required by the user on the user's matching mobile edge computing server.

[0016] Under the premise of meeting the edge service hit rate requirements and service delay requirements, the present invention aims to maximize the data transmission rate of mobile users and minimize the service pre-deployment delay, and realizes edge service pre-deployment that can predict user paths and realize dynamic matching between MEC servers and users. On the one hand, the user's historical trajectory is deeply analyzed to obtain the user's predicted location. For example, the spatiotemporal information and shape characteristics of the user's trajectory can be used for prediction to obtain a more accurate location prediction result. On the other hand, by realizing bilateral stable matching between users and MEC servers, and pre-deploying services on MEC servers near the predicted location, the task processing delay can be reduced and the user experience can be improved.

[0017] Furthermore, the specific measurements of trajectory similarity include:

[0018] The inter-segment slope of the trajectory is used to measure the similarity of the trajectory shape distance, and the shape distance between the trajectories is obtained;

[0019] The Euclidean distance is used to calculate the trajectory position distance between trajectories;

[0020] The measure of trajectory similarity between trajectories is obtained based on the weighted sum of the shape distance and the trajectory position distance between trajectories.

[0021] Furthermore, constructing the trajectory transition probability matrix of the current user based on the filtered user historical trajectory specifically includes:

[0022] Obtain a similar trajectory set based on the filtered user historical trajectories;

[0023] The trajectory point set covered by the similar trajectory set is used as the trajectory key point set;

[0024] Traverse each trajectory key point, count the trajectory transfer probability of the user moving from the current trajectory key point to each trajectory key point in the similar trajectory set, and obtain the trajectory transfer probability matrix of the current user.

[0025] Furthermore, the trajectory transition probability of the user moving from the current trajectory key point to each trajectory key point is set as:

[0026] Traverse each trajectory key point in the trajectory key point set, and count the number of times the user moves from the current trajectory key point to each trajectory key point in the similar trajectory set , where e is the current trajectory key point index, E is the trajectory index where the current trajectory key point e is located, and f is the trajectory key point index currently moved to;

[0027] According to the formula Calculate the trajectory transfer probability corresponding to the current trajectory key point e to the current trajectory key point f ,in, represents the set of trajectory key points, Represents a set of trajectory key points Any trajectory key point in .

[0028] Furthermore, when predicting the predicted position of each user based on the trajectory transition probability matrix of each user, a Markov model is used to predict the predicted position of each user.

[0029] Furthermore, when obtaining the bilateral matching result of the user-mobile edge computing server in the candidate mobile edge computing server set, the total transmission efficiency of all current users is: the sum of the user achievable data transmission rates of all matched users, where the user achievable data transmission rate is:

[0030] ;

[0031] Where, subscript i is the index of the edge computing server, subscript j is the user index, B is the channel bandwidth, is the user's noise power, is the path loss between user j and edge computing server i, is the transmission power of user j accessing edge computing server i, The transmission power of user j accessing other edge computing servers except edge computing server i.

[0032] Furthermore, when obtaining the bilateral matching result between the user and the mobile edge computing server in the candidate mobile edge computing server set, the matching method adopts a many-to-many bilateral matching strategy.

[0033] Furthermore, when pre-deploying the services required by the user on the user's matching mobile edge computing server, satisfying the user's service request means that the total delay of pre-deploying the services required by the user does not exceed the time limit specified by the user, wherein the total delay of pre-deploying the services required by the user is the sum of all transmission delays and deployment delays.

[0034] Further, when the service required by the user is pre-deployed on the user's matching mobile edge computing server, the server matching optimization target solution is performed based on the service request initiated by the user, wherein the service request initiated by the user includes the pre-deployed service , Pre-deployment service Required packet size and pre-deployment services Time Limit , where subscript j is the user index.

[0035] Furthermore, when pre-deploying services required by users on the user's matching mobile edge computing server, the mean field game strategy is used to execute the service pre-deployment optimization goal to obtain the service pre-deployment results of each user.

[0036] In summary, the technical solution provided by the present invention brings at least the following beneficial effects:

[0037] (1) Low latency. First, the present invention pre-deploys services on a server closer to the user based on the user's predicted location, which can reduce network transmission time and thus reduce service latency. Second, the present invention can pre-deploy services to load relevant content or data onto the server before the user's request arrives, thereby achieving faster service response speed and increasing user stickiness. Finally, through the multi-MEC server matching process and the pre-deployment service process, the load can be distributed to different MEC servers, reducing the pressure on a single server and improving the stability and reliability of the overall system.

[0038] (2) High transmission rate. The present invention can predict the user's future location based on the mobility pattern corresponding to the user's historical trajectory and current trajectory, so as to cache the data corresponding to the deployed service in advance on the MEC server near the user. In this way, when the user really needs to access the data, he can directly obtain it from the nearby MEC server, which disperses the load of data transmission, avoids excessive congestion in a single area, and avoids waiting and congestion during data transmission, thereby improving bandwidth utilization.

[0039] (3) High cost-effectiveness. Location prediction can predict the user's location information in advance, allowing service providers to allocate resources and deploy services more effectively to avoid resource waste. Pre-deployment services can also prepare the required equipment and materials in advance based on the predicted location information, thereby avoiding additional costs caused by temporary resource deployment and reducing additional costs caused by long waiting times.

[0040] (4) Strong applicability. The present invention takes into account the dynamic deployment of services under multi-dimensional resources, and can also adaptively adjust the deployment strategy in combination with the reward function configured by the mean field game strategy. In addition, the present invention is user-centric and can provide customized services for different services of different users. DETAILED DESCRIPTION

[0041] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described in detail and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0042] With the popularization of mobile devices and edge computing technology, users are becoming more and more mobile. Service pre-deployment based on location prediction can dynamically adjust the deployment location of service resources according to the user's movement trajectory and location changes to meet the user's real-time needs. In addition, with the continuous emergence of new services, the requirements for service quality and response speed are getting higher and higher. Efficient service pre-deployment solutions can provide strong support for new business models and promote business innovation and development. Therefore, the accuracy of user location prediction and the efficiency of service pre-deployment solutions are important links in promoting the development of edge networks.

[0043] In response to the above application requirements, an embodiment of the present invention provides an edge service pre-deployment method based on location prediction. Under the premise of meeting the edge service hit rate requirements and service delay requirements, this method aims to maximize the data transmission rate of mobile users and minimize the service pre-deployment delay, and proposes an edge service pre-deployment that can predict user paths and achieve dynamic matching between MEC servers and users. On the one hand, the embodiment of the present invention conducts an in-depth analysis of the user's historical trajectory to predict the user's location. Specifically, the spatiotemporal information and shape features of the user's trajectory can be used to predict the location to obtain a more accurate location prediction result. On the other hand, the embodiment of the present invention reduces the task processing delay and improves the user experience by achieving bilateral stable matching between the user and the MEC server and pre-deploying the service on the MEC server near the predicted location.

[0044] An edge service pre-deployment method based on location prediction provided by an embodiment of the present invention can be applied to a communication system architecture with a three-layer architecture of "end-edge-cloud", wherein the edge layer is composed of a macro base station and M micro base stations, and an MEC server can be deployed on the base station (macro base station and micro base station). Among them, the macro base station is connected to the core network, and a central controller is configured on it, which is responsible for the working state switching of the micro base station, and the micro base station manages the MEC server under its jurisdiction according to the macro base station signal; the user accesses the edge network through the micro base station or the macro base station, and the micro base station accesses the macro base station through the backhaul network. The user can switch between various base stations according to his own needs. Under this communication system architecture, the embodiment of the present invention first predicts the user's location using the spatiotemporal information and shape information of the historical trajectory, and then completes the bilateral stable matching between the user and the MEC server with the goal of maximizing the user's achievable data transmission rate on this basis, and finally models the service required for the user to offload the task as a mean field game model, so as to achieve accurate user location prediction and efficient service pre-deployment scheme under the premise of meeting the user's data transmission rate and service delay.

[0045] The edge service pre-deployment method based on location prediction provided by the embodiment of the present invention mainly includes three stages: user location prediction stage, user-centric multi-server matching stage and matching server-based service pre-deployment stage. Among them, the processing of the user location prediction stage includes: collecting the user history trajectory of the user to obtain the user history trajectory set of each user; filtering out the user history trajectory whose trajectory similarity with the user's current trajectory is greater than or equal to the similarity threshold from the user history trajectory set of each user, and constructing the trajectory transfer probability matrix of the current user based on the filtered user history trajectory; predicting the predicted location of each user based on the trajectory transfer probability matrix of each user. The user-centric multi-server matching stage is: based on the predicted location of each user, for each MEC server to be deployed, if the current user is within the communication range of the current MEC server, then the mobile edge computing server is added to the candidate MEC server set of the current user; under the premise of meeting the upper limit of the MEC server that a single user can match and the upper limit of the number of users that a single MEC server can serve, the total transmission efficiency of all current users is maximized as the server matching optimization goal, and the bilateral matching result of the user-MEC server is obtained in the candidate MEC server set to obtain the matching MEC server set of each user. The service pre-deployment stage based on the matching server is: on the premise of meeting the user's service request, the service pre-deployment optimization target is set with the minimum service pre-deployment delay, and the service required by the user is pre-deployed on the user's matching MEC server.

[0046] Taking the three-layer network architecture (end-edge-cloud) as an example, the specific implementation process of an edge service pre-deployment method based on location prediction provided by an embodiment of the present invention is described as follows:

[0047] Define q to represent the number of users included in the user layer, then the corresponding user set can be expressed as , there is a macro base station at the edge layer and Compared with macro base stations, micro base stations have smaller transmission power and coverage, and are more suitable for deploying edge servers to provide edge services. Therefore, the embodiment of the present invention considers deploying MEC servers on micro base stations, where the MEC server set can be expressed as .

[0048] 1. User (mobile user) location prediction.

[0049] Assume user From the base station Move to base station , , you need to add some users in advance The services required for the task are coordinated to and This ensures that users can offload the corresponding tasks to the base station ahead for calculation in advance, reducing the delay in service pre-deployment and task processing result transmission. Since the user's next location is only related to the current location, in the embodiment of the present invention, the trajectory transfer probability matrix is ​​established using the historical trajectory of users with high similarity, and the user's location at a future time point is predicted through the Markov model to obtain the user's predicted location.

[0050] Assume that the user's trajectory sampling point set is ,in Indicated in The sampling position of the user trajectory at time , and the user makes a uniform linear motion between any two adjacent sampling points. In order to further reduce the usage of storage resources, the sampling point set can also be clustered by hierarchical clustering (HAC). Perform hierarchical division to achieve trajectory clustering, thereby reducing storage capacity and unifying trajectory dimensions. That is, the sampling points of each trajectory obtained after trajectory clustering are regarded as key points, and the corresponding trajectory can be expressed as: , where p represents the number of key points (sampling points after clustering) included in each clustered trajectory.

[0051] In order to select suitable historical trajectories as samples for predicting user locations, it is necessary to measure the similarity between historical trajectories and current trajectories. The steps are as follows:

[0052] Step S1-1: Construct synchronization trajectory.

[0053] definition , Respectively represent the user's current trajectory and any historical trajectory. The current trajectory and historical trajectory can be expressed as: , , Among them, e, The trajectory points (sampling points or the number of sampling points after clustering) included in the current trajectory and the historical trajectory respectively. , Each sampling moment in the union of the sampling moments of the trajectory points ,like There are track points ,but If there is no trajectory point, then Insert new track points (e.g., based on mean interpolation to obtain its value), and vice versa. Thus, a new synchronous trajectory is obtained and . That is, traversal trajectory , The sampling moments of the key points of the two trajectories are combined to form a sampling moment, and each sampling moment of the set is used to detect whether there are trajectory points in the two trajectories, and the missing trajectory points are supplemented by interpolation to obtain a new synchronous trajectory. and .

[0054] Step S1-2: Calculate the trajectory shape distance.

[0055] The inter-segment slope of the trajectory is used as a metric for the similarity of trajectory shape distance. Specifically, the synchronous trajectory and Shape distance The definition is as follows:

[0056] (1)

[0057] in, , Synchronous trajectory and The number of trajectory points, Synchronous trajectory and The The location of the trajectory point.

[0058] Step S1-3: Calculate the trajectory position distance.

[0059] The synchronous trajectory can be obtained by calculating the Euclidean distance and Location distance for:

[0060] (2)

[0061] Step S1-4: Calculate the trajectory distance.

[0062] The trajectory similarity is inversely proportional to the trajectory distance. According to the trajectory shape distance and trajectory position distance, the trajectory distance can be obtained as follows:

[0063] (3)

[0064] in, The preset weights.

[0065] Step S1-5: Filter similar historical trajectories.

[0066] Given a similarity threshold ,when When and Similar, and the historical trajectory Add to the current user's similar trajectory collection middle.

[0067] Next, to predict the trajectory The next trajectory point needs to be combined with a set of similar trajectories The set of covered trajectory points constructs the corresponding trajectory transition probability matrix :

[0068] 1) Group similar trajectories The covered trajectory points are regarded as the trajectory key point set, and the symbol Represents, and n represents the set of trajectory key points The number of key points of the trajectory;

[0069] 2) Traverse the trajectory key point set For each trajectory key point in , will be collected Each trajectory key point in is used as a trajectory key point The candidate point of the next predicted trajectory point is defined as the candidate point , and then count the number of users from the key points of the trajectory in the similar trajectory set Move to each candidate point separately The number of moves is ;

[0070] Then calculate the trajectory key points according to the trajectory transfer probability calculation formula To the candidate point The trajectory transition probability :

[0071] (4)

[0072] 3) Based on the calculated trajectory transfer probability , construct the corresponding The trajectory transition probability matrix of dimension is recorded as , whose expression is:

[0073] (5)

[0074] Next, define Indicates the user's current location. is the predicted location of the user, then:

[0075] (6)

[0076]

[0077] in, Represents the trajectory transition probability matrix Get the current location The moving position point corresponding to the maximum trajectory transfer probability, that is, the predicted trajectory point found, Represents the predicted trajectory point With current location The conditional probability function of .

[0078] In the embodiment of the present invention, after calculating the trajectory transition probability matrix After that, the Markov model can be used to predict the user's location at a future time point, thereby obtaining the user's predicted location.

[0079] 2. User-centric multi-server matching.

[0080] After the user location prediction stage, the user's predicted location is obtained, and then the service matching the user's task is deployed in advance on the MEC server, which is deployed on a macro base station or a micro base station. Therefore, a user-centric bilateral matching of the user and the MEC server is required to maximize the user data transmission rate.

[0081] Consider the user's arrival at the predicted location When the MEC server is set to 1, the user can dynamically select MEC servers on multiple base stations near him to create a MEC server cluster, and each user can select up to MEC servers, accordingly, each MEC server can be associated with up to users, of which , The value is the user configuration value.

[0082] In one embodiment, the specific steps of performing user-centric bilateral matching between a user and a MEC server in the embodiment of the present invention include:

[0083] Step S2-1: Build a set of candidate MEC servers.

[0084] When the user On MEC server Communication range If the server is within the user A collection of alternative MEC servers In ,in, Indicates user and MEC server The distance between.

[0085] Step S2-2: Calculate the user's achievable data transmission rate.

[0086] Defining indicator functions Indicates user and MEC server If there is a matching relationship between them, then:

[0087] (7)

[0088] Assume access to MEC server All users of the system share the power equally, so the user Connect to the MEC server The transmission power for:

[0089] (8)

[0090] in, Indicates MEC server Power of the base station.

[0091] user The power received from other base stations is:

[0092] (9)

[0093] in, To characterize the user Access to the MEC server The transmission power of other MEC servers outside is the index of the candidate MEC server, and , Indicates the set of candidate MEC servers The MEC servers, For users Connect to the MEC server transmission power.

[0094] definition is the reference distance of path loss, it can be known that user and MEC server The path loss between for:

[0095] (10)

[0096] in, and denote the constant coefficient and path loss exponent depending on the antenna characteristics and the average channel loss, respectively.

[0097] Based on path loss , transmission power and other transmission powers Available as user Connect to the MEC server When the user The achievable data transfer rate is:

[0098] (11)

[0099] in, is the channel bandwidth, is the noise power of the user, which is specifically the additive white Gaussian noise power in this example.

[0100] Step S2-3: Modeling of matching optimization problem.

[0101] In order to maximize the achievable total system transmission rate, the following MEC server matching optimization problem can be constructed:

[0102] (12)

[0103] (12-1)

[0104] (12-2)

[0105] (12-3)

[0106] (12-4)

[0107] Among them, constraint (12-1) is an indicator function constraint, constraint (12-2) indicates that the user matched with the MEC server must be within the communication range of the MEC server, and constraint (12-3) indicates that in the same time slot, a user can match at most MEC servers, constraint (12-4) means that in the same time slot, a MEC server can be at most Provide services to users.

[0108] The server matching optimization problem shown in formula (12) is a non-convex problem. Due to the complexity of non-convex optimization problems and the characteristics of multiple local optimal solutions, its optimal solution is usually difficult to find in polynomial time. Therefore, in the embodiment of the present invention, the optimization problem is modeled as a many-to-many bilateral matching problem for solution.

[0109] Assume that all users and MEC servers receive macro base stations Instructions, macro base station After giving multiple bilateral matching schemes, users and MEC servers can choose matching schemes according to their different preferences. The specifics of the many-to-many bilateral matching problem are:

[0110] First, create a MEC server preference list and a user preference list.

[0111] Known macro base stations The matching relationship given ,definition Indicates MEC server Given a descending preference vector of users, where each user's preference value is The definition is as follows:

[0112] (13)

[0113] (14)

[0114] in, Indicates that based on matching relationship The obtained data is currently connected to the MEC server The user collection on Indicates that you are currently connected to the MEC server Users on Currently connected to the MEC server The user index on .

[0115] In matching relationship Next, user Descending preference vector for MEC servers The preference value of each MEC server is defined as:

[0116] (15)

[0117] in, For macro base stations The matching relationship given With the current user The index of the matching MEC server.

[0118] From the preference list, we can see that the descending preference vector and It is related to the power received by the user, so the user's preference is not only related to the matched MEC server, but also to other users matched to the same base station, and vice versa. In other words, any single pair of matching relationships will be affected by itself and the matching relationships of other pairs. Therefore, when creating a preference list, it is necessary to consider the global matching results, and this matching process is called a matching game under a peer relationship.

[0119] Next, a swap match is performed.

[0120] In order to better describe the dependency relationship between users and MEC servers and find stable matching results, it is assumed that users and If the number of initial matching servers of is greater than 0, then the user and The exchange match is defined as follows:

[0121] (16)

[0122] Among them, users and MEC server is a pair of initial matching objects, user and MEC server is another pair of initial matching objects, Indicates exchange user and Matchmaking server.

[0123] Define interlocking pairs in swap matching. When the user and When the interlock pair is on, it means the user and When exchanging matches, the preference value of at least one party will increase, and the constraints are as follows:

[0124] Constraint 1: For users and For any matching relationship in the , if the matching is exchanged, the new preference value is not less than the initial preference value;

[0125] Constraint 2: User and There is at least one pair of matching relationships in , so that the new preference value is greater than the initial preference value.

[0126] According to the definition of exchange matching and interlocking pair, it can be known that when the macro base station The given match When there is no interlocking pair in Only then can the two-way exchange be stable.

[0127] In this embodiment, the specific process of executing the many-to-many bilateral matching strategy to achieve the server matching optimization goal is as follows:

[0128] Assuming macro base station Given an initial match ,and Satisfy the constraints in the matching optimization problem. In this case, each user sends an interlocking pair verification to the MEC server in the candidate MEC server set. If there is an interlocking pair between the initiating user and the initial matching user of the MEC server (i.e., the above two constraints are met), an exchange match is performed. The above process is repeated until there are no interlocking pairs in the system. At this time, the matching relationship is a bilateral stable match between the user and the MEC server. .

[0129] At this point, based on the user's predicted location, the matching relationship between the user's predicted location and multiple MEC servers is obtained under the premise of maximizing the user's rate.

[0130] 3. Pre-deployment of services based on matching servers.

[0131] After obtaining the optimal matching result between the user and the MEC server, the user A collection of alternative MEC servers Update to match set , and the user The services required for the offloading task are deployed in advance to On the MEC servers included, that is, under the premise of satisfying the user's service request, the service deployment optimization goal is to minimize the service pre-deployment delay. On the user's matching MEC server set (i.e. ) The corresponding matching MEC server pre-deploys the services required by the user to ensure that when the user arrives at the predicted location When the edge service is connected, the user can obtain edge services in time, reduce service delays, and further improve the user's quality of service (QoS).

[0132] Set up MEC server The available resources are ,in Represents storage resources. represents computing resources, Indicates a transport resource.

[0133] when When MEC server The microservices deployed on , Indicates deployment on MEC server The resources required by the hth microservice on the system, and the same microservice in the system Required Resources same, Microservices The required storage resources, computing resources, and transmission resources; N represents the number of microservices, and the microservice index .

[0134] In time slot t, the service that the system needs to deploy in advance is ,in Indicates user The services required in the future are composed of a series of microservices and can be expressed as:

[0135] (17)

[0136] in, represents the hth microservice of the system, Indicates that on the MEC server Medium Service The required set of microservices.

[0137] In order to better represent the dependency relationship between microservices, users, and MEC servers, two indicator variables are defined here. and :

[0138] (18)

[0139] (19)

[0140] In one embodiment, the specific steps of obtaining the service deployment optimization target based on the above configuration include:

[0141] Step S3-1: Create a service request model.

[0142] Since the service needs to be at the predicted location when the user arrives Pre-deployed, so users There are strict time limits when requesting services. The service request can be expressed as:

[0143] (20)

[0144] in, Represents a pre-deployed service The required packet size, Represents a pre-deployed service time limit, and the time limit is determined by the user The speed and channel conditions determine it.

[0145] Step S3-2: Calculate the service pre-deployment delay.

[0146] The service pre-deployment delay consists of the transmission delay of the service migration to the corresponding MEC server and the pre-deployment delay on the MEC server.

[0147] On the MEC server and MEC server Transfer between users Required services Delay For transmission services The sum of the latency of the required microservices is expressed as:

[0148] (twenty one)

[0149] Pre-deployment users Required services Delay Pre-deploy services on matching servers The sum of the latencies of the required microservices:

[0150] (twenty two)

[0151] Therefore, pre-deployment users Required services The total delay is the sum of all transmission delays and deployment delays:

[0152] (twenty three)

[0153] Step S3-3: Establish a service pre-deployment optimization problem.

[0154] In order to meet the user's service request and minimize the service pre-deployment delay, this embodiment proposes the following service pre-deployment optimization problem:

[0155] (twenty four)

[0156] (24-1)

[0157] (24-2)

[0158] (24-3)

[0159] (24-4)

[0160] (24-5)

[0161] Step S3-4: Create a mean field game model.

[0162] In order to solve the service deployment optimization problem shown in formula (24), this embodiment models it as a mean field game (MFG) problem and can be solved by reinforcement learning.

[0163] Without loss of generality, the user Required services For example, the service will be The required set of microservices is modeled as a directed graph , which constitutes The microservices are a directed graph If there is a communication link between the MEC servers where the microservices are located Then it forms a directed graph , and denote the edge set as c.

[0164] Step S3-4-1: Establish a microservice distribution model.

[0165] At time slot t, let the MEC server Deployed microservices The number of , then the distribution of microservices in the system It can be expressed as:

[0166] (25)

[0167] Step S3-4-2: Establish a service pre-deployment probability transfer matrix.

[0168] make represents the transition probability matrix of the updated microservices in the system at time slot t. Line Column value Indicates that microservices Replace with microservices probability.

[0169] Step S3-4-3: Calculate the reward function.

[0170] According to the transfer vector (Transition probability matrix of microservices The row vector of the system) and the distribution of microservices in the system , you can deploy microservices The reward function If and only if microservices Execute an action You will get the reward when:

[0171] (26)

[0172] in, Indicates that microservices Replace with microservices The reward value is the preset value.

[0173] Step S3-4-4: Design the forward-backward equation.

[0174] According to the distribution of microservices, the backward equation (HJB equation) is:

[0175] (27)

[0176] And the forward equation (FPK equation) is:

[0177] (28)

[0178] in, Indicates the distribution of microservices in the next time slot within the system, represents the value function obtained based on the distribution of microservices in the system at time slot t, that is, the optimal distribution of microservices can be solved by the HJB equation. The FPK equation represents the evolutionary process of the mean field game between microservices.

[0179] At this point, the value function is updated iteratively multiple times with the goal of maximizing the reward function within the system. , the best reward and the optimal deployment plan of the microservice can be obtained. In the embodiment of the present invention, the iteration convergence condition can be set to reach a preset maximum number of iterations or the value convergence of the value function.

[0180] In an edge service pre-deployment method based on location prediction provided by an embodiment of the present invention, the shape information and spatiotemporal information of the user trajectory are used to judge the trajectory similarity, a path set that meets the similarity standard is screened out from the user's historical data, and a transition probability matrix is ​​established to predict the user's location at the next time point, which greatly increases the accuracy of user location prediction while ensuring the service time. With the user as the center, a preference list between the user and the MEC server is established, a bilateral stable match between the user and the MEC server is achieved at the predicted location, and the service pre-deployment problem is modeled as a mean field game model, which increases the user data transmission rate while reducing the service pre-deployment delay.

[0181] When applying a method for pre-deploying edge services based on location prediction provided by an embodiment of the present invention, it is first necessary to configure an execution program for executing the user location prediction provided by the embodiment of the present invention on an edge device (such as a router, switch or IoT gateway, etc.); then, connect these edge devices to a cloud server or a central data center through a network protocol (such as a message queue telemetry transmission protocol MQTT or a constraint application protocol CoAP, etc.); in addition, the user's real-time location data can be collected by using location-based sensors (such as a global positioning system GPS or a radio frequency identification RFID, etc.) to further improve the accuracy of the user location prediction of the embodiment of the present invention. Finally, the edge device can manage and monitor the pre-deployment of user request services through the edge computing platform, that is, the user-centric user-MEC server bilateral matching provided by the embodiment of the present invention and the service pre-deployment based on the matching server can be performed through the edge computing platform to achieve faster data processing and response.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0183] The above are only some embodiments of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the creative concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for pre-deploying edge services based on location prediction, characterized in that: The following steps are involved: Collect the user's historical trajectory and obtain the user's historical trajectory set for each user; Filter out the user historical trajectories whose trajectory similarity with the user's current trajectory is greater than or equal to the similarity threshold from the user historical trajectory set of each user, and construct the trajectory transition probability matrix of the current user based on the filtered user historical trajectories; Among them, the specific measurements of trajectory similarity include: Definition S E , S F Represent the user's current trajectory and any user's historical trajectory respectively, traversing the trajectory S E , S F The sampling moments of the trajectory points of the two trajectories are combined to form the sampling moments of the two trajectories. The missing trajectory points are supplemented by interpolation to obtain the new synchronous trajectory S′. E and S′ F ; The inter-segment slope of the trajectory is used to measure the similarity of the trajectory shape distance, and the shape distance between the trajectories is obtained. Among them, k and k′ are the synchronous trajectory S′ E and S′ F The number of trajectory points, They are the synchronous trajectories S′ E and S′ F The position of the τth trajectory point; The Euclidean distance is used to calculate the trajectory position distance between trajectories. The measure of trajectory similarity between trajectories is obtained based on the weighted sum of the shape distance and trajectory position distance between trajectories. Among them, ε is the preset weight; Predicting the predicted position of each user based on the trajectory transition probability matrix of each user; Based on the predicted location of each user, for each mobile edge computing server to be deployed, if the current user is within the communication range of the current mobile edge computing server, the mobile edge computing server is added to the set of candidate mobile edge computing servers for the current user; Under the premise of satisfying the upper limit of edge computing servers that a single user can match in the same time slot and the upper limit of the number of users that a single edge computing server can serve, the server matching optimization goal is to maximize the total transmission efficiency of all current users, obtain the bilateral matching results of users and mobile edge computing servers in the set of candidate mobile edge computing servers, and obtain the matching mobile edge computing server for each user; and when obtaining the bilateral matching results of users and mobile edge computing servers in the set of candidate mobile edge computing servers, the matching method adopts a many-to-many bilateral matching strategy, and all users and MEC servers receive macro base stations. Instructions, macro base station After giving multiple bilateral matching schemes, users and MEC servers select matching schemes according to their different preferences; The preference value α of each user ij (μ) is defined as: a i1 (μ)≥a ij (m)≥a iq (m) Where, 1≤i≤w,1≤j≤q, w represents the number of MEC servers, q represents the number of users, R i,j Represents user u j Access to MEC server m i The user can reach the data transmission rate at Indicates the currently connected MEC server n based on the matching relationship μ i The user set on ,u r′ Indicates that the MEC server is currently connected. i r′ is the user currently connected to the MEC server m i User index on ; The preference value β of each MEC server hj (μ) is defined as: β h,j (μ) = R h,j , where h is the macro base station The given matching relationship μ and the current user u j The matching MEC server index; Under the premise of satisfying the user's service request, the service required by the user is pre-deployed on the user's matching mobile edge computing server with the minimum service pre-deployment delay as the service pre-deployment optimization target; wherein satisfying the user's service request means that the total delay of pre-deploying the service required by the user does not exceed the time limit specified by the user, and the total delay of pre-deploying the service required by the user is the sum of all transmission delays and deployment delays; when pre-deploying the service required by the user on the user's matching mobile edge computing server, the mean field game strategy is adopted to execute the service pre-deployment optimization target; based on the service request initiated by the user, the server matching optimization target is solved, wherein the service request initiated by the user includes the pre-deployed service x j , Pre-deployment service j Required packet size ∈ j and pre-deployment services j Time Limit Subscript j is the user index; The expression of the service pre-deployment optimization objective is: Among them, time j Indicates the pre-deployed user u j Required service x j The total delay is the sum of all transmission delays and deployment delays. Represents a matching set, Υ i Indicates MEC server m i Available resources, i ={e i ,c i ,v i },e i Indicates storage resources, c i represents computing resources, v i Represents transmission resources; microservice Θ i ={θ i,1 ,…,θ i,N }, θ i,h Indicates deployment on MEC server m i The resources required by the hth microservice on the system, and the same microservice in the system Required Resources same, Microservices The required storage resources, computing resources, and transmission resources; N represents the number of microservices, and the microservice index Two indicator variables The expressions are: Among them, the optimization goals of service pre-deployment using the mean field game strategy include: (1) At time slot t, let MEC server m i Deployed microservices The number of Distribute microservices within the system It is expressed as: (2) Order represents the transition probability matrix of the microservice updated in time slot t, where Line Column value Indicates that microservices Replace with microservices probability; (3) The transition probability matrix of microservices The row vector of is defined as the transfer vector According to the transfer vector and distribution Deploy microservices The reward function is: in, Indicates that microservices Replace with microservices The reward value, the preset value; (4) The backward equation is obtained based on the microservice distribution: And the forward equation is: in, Indicates the distribution of microservices in the next time slot within the system, represents the value function obtained based on the distribution of microservices in the system at time slot t, and the optimal distribution of microservices is solved by the backward equation; To maximize the reward function within the system, the value function is updated iteratively multiple times Get the best rewards and the best deployment plan for microservices.

2. The edge service pre-deployment method based on location prediction according to claim 1, characterized in that: Constructing the trajectory transition probability matrix of the current user based on the filtered user historical trajectory specifically includes: Obtain a similar trajectory set based on the filtered user historical trajectories; The trajectory point set covered by the similar trajectory set is used as the trajectory key point set; Traverse each trajectory key point, count the trajectory transfer probability of the user moving from the current trajectory key point to each trajectory key point in the similar trajectory set, and obtain the trajectory transfer probability matrix of the current user.

3. The edge service pre-deployment method based on location prediction according to claim 2, characterized in that: The trajectory transition probability of the user moving from the current trajectory key point to each trajectory key point is set as: Traverse each trajectory key point in the trajectory key point set, and count the number of times the user moves from the current trajectory key point to each trajectory key point in the similar trajectory set Among them, e is the index of the current trajectory key point, E is the trajectory index where the current trajectory key point e is located, and f is the index of the trajectory key point currently moved to; According to the formula Calculate the trajectory transfer probability corresponding to the current trajectory key point e to the current trajectory key point f in, represents the set of trajectory key points, Represents a set of trajectory key points Any trajectory key point in .

4. The edge service pre-deployment method based on location prediction according to any one of claims 1 to 3, characterized in that: When predicting the predicted position of each user based on the trajectory transition probability matrix of each user, a Markov model is used to predict the predicted position of each user.

5. The edge service pre-deployment method based on location prediction according to claim 1, characterized in that: When obtaining the bilateral matching result of the user and the mobile edge computing server in the candidate mobile edge computing server set, the total transmission efficiency of all current users is: the sum of the user achievable data transmission rates of all matched users, where the user achievable data transmission rate is: Where, subscript i is the index of the edge computing server, subscript j is the user index, B is the channel bandwidth, σ 2 is the user's noise power, μ i,j is the path loss between user j and edge computing server i, P i,j is the transmission power of user j accessing edge computing server i, The transmission power of user j accessing other edge computing servers except edge computing server i.

Images