Mobile edge caching system and method based on user mobility and preference

Abstract

The invention relates to the technical field of mobile edge computing, in particular to a mobile edge caching system and method based on user mobility and preference, and the method comprises the following steps: constructing a mobile edge caching system; the mobile edge caching system finds a base station for providing service for a user according to the movement of the user based on an algorithm; using collaborative filtering to predict interested contents of the user to obtain predicted contents; according to the method, the mobility and preference of the user are comprehensively considered, the content which the user is interested in is cached on the optimal edge node, the delay is reduced to the maximum extent, the cache hit rate is increased, and the user experience is improved. According to the method, a neural network collaborative filtering algorithm is utilized to predict the content interested by the user, it is guaranteed that prediction has the characteristics of good accuracy, real-time performance and low calculation complexity, the user interest is determined through user preference and serves as the content to be requested by the user, and the optimal edge node providing service for the user is calculated in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile edge computing, and in particular, to a mobile edge caching system and method based on user mobility and preferences. Background Art

[0002] With the continuous development of science and technology, cutting-edge technology applications such as autonomous driving, robots, virtual reality, and live broadcasting have emerged in people's lives. While providing convenience for people, they have also generated a large amount of data, exacerbating the pressure on the backhaul link. To solve this problem, mobile edge caching technology has emerged. By providing a storage function at the network edge, services are provided to users nearby, thereby reducing the pressure on the backhaul link, improving transmission efficiency, and enhancing the user experience. However, users are mobile, so service requests are relatively random, and the storage space of edge nodes is limited and cannot cache all the files that users may need. To improve the cache hit rate of edge nodes, the problem of where to cache what content is particularly important.

[0003] In the prior art, the literature: N. Li, L. Zhai, S. Song, X. Zhu, Y. Li, and F. Yang, "A Cache Allocation Strategy Using Reinforcement Learning Based on User Preferences in Mobile Edge Computing", Concurr. Comput. Pract. Exp, vol. 36, pp. e7991, 2024. proposed a cache resource allocation and cache space adjustment strategy suitable for edge computing systems, which greatly improved the cache hit rate. However, this strategy only considered the impact of user preferences, that is, it solved the problem of "what content to cache", and did not consider the impact of user mobility on the cache hit rate. Summary of the Invention

[0004] The purpose of the present invention is to provide a mobile edge caching system and method based on user mobility and preferences, aiming to comprehensively consider user mobility and preferences to improve the cache hit rate, thereby reducing the edge cache of the backhaul link load.

[0005] To achieve the above object, in the first aspect, the present invention provides a mobile edge caching method based on user mobility and preferences, including the following steps:

[0006] Construct a mobile edge caching system;

[0007] The mobile edge caching system finds the base station that provides services for it according to user mobility based on an algorithm;

[0008] Use collaborative filtering to predict the content that the user is interested in to obtain predicted content;

[0009] Cache the predicted content on the edge server of the base station that provides services for the user.

[0010] Among them, the specific method by which the mobile edge caching system finds the base station that provides services for a user according to the user's movement based on an algorithm is as follows:

[0011] Establish a mathematical model of edge nodes;

[0012] Based on the mobile edge caching system and the mathematical model, construct a solution, and use an improved differential evolution algorithm to solve for the base station that provides services for the user.

[0013] Among them, the specific method of using collaborative filtering to predict the content that a user is interested in and obtaining the predicted content is as follows:

[0014] Analyze and obtain information about the files downloaded by the user from the user's previous download history records to generate a training set and a test set;

[0015] Based on the training set and the test set, train and test the collaborative filtering model, and use the collaborative filtering model to predict the content that the user is interested in.

[0016] Among them, the collaborative filtering model includes an output layer, a NeuralCF layer, and an input layer. Among them, the output layer is the input of the NeuralCF layer, and the NeuralCF layer is the input of the input layer.

[0017] In a second aspect, the present invention also provides a mobile edge caching system based on user mobility and preferences, which is applied to the mobile edge caching method based on user mobility and preferences as described in the first aspect above, and is characterized in that;

[0018] It includes a cloud server, a small base station, a macro base station, and a server, and the cloud server, the macro base station, the server, and the small base station are connected in sequence.

[0019] A mobile edge caching system and method based on user mobility and preferences of the present invention constructs a mobile edge caching system. The mobile edge caching system finds the base station that provides services for a user according to the user's movement based on an algorithm, uses collaborative filtering to predict the content that the user is interested in, obtains the predicted content, and caches the predicted content on the edge server of the base station that provides services for the user. This method combines the actual situation, comprehensively considers the user's mobility and preferences, caches the content that the user is interested in on the optimal edge node, thereby minimizing the delay and increasing the cache hit rate to the greatest extent. The optimal edge node is solved by using an improved differential evolution algorithm, and the prediction of the content that the user is interested in is realized by using a neural network collaborative filtering algorithm. Through offline model training and online prediction, it is ensured that the prediction has good accuracy, real-time performance, and low computational complexity characteristics. The user's interest is determined through the user's preferences and used as the content that the user is likely to request. The best edge node that provides services for the user is calculated in advance, and the content is cached in advance, improving the service quality. Brief Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 is a schematic diagram of a mobile edge caching system based on user mobility and preferences provided by the present invention.

[0022] Figure 2 is a schematic flow chart of a mobile edge caching method based on user mobility and preferences provided by the present invention.

[0023] Figure 3 is a NeuralCF model diagram.

[0024] Figure 4 is a flow chart of a mobile edge caching method based on user mobility and preferences provided by the present invention.

[0025] Figure 5 is a flow chart of the specific method for a user to move based on the mobile edge caching system to find a base station that provides services for it.

[0026] Figure 6 is a flow chart of the specific method for using collaborative filtering to predict the content of interest to users and obtain the predicted content.

[0027] In the figure: 1 - cloud server, 2 - small base station, 3 - macro base station, 4 - server. Detailed Embodiments

[0028] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.

[0029] Please refer to Figures 2 to 6 , in the first aspect, the present invention provides a mobile edge caching method based on user mobility and preferences, including the following steps:

[0030] S1 Build a mobile edge caching system;

[0031] S2 The mobile edge caching system finds a base station that provides services for it according to the user's movement based on an algorithm;

[0032] Specific method:

[0033] S21 Establish a mathematical model of the edge node;

[0034] In the embodiment of the present invention, the connectivity between the user Um and the N small base stations 2 can be represented by the matrix Cm, and where represents the connectivity between the user Um and the base station s. If it means that the user Um is not connected to the base station s; if it means that the user Um is connected to the base station s. The total file Ff = {1, 2,..., f,..., F}, and the size of each file is Size = {Size1, Size2,..., Sizef,..., SizeF}. The files stored in the edge server 4 of the base station s are If it means that file 1 is stored on the edge server 4 of the base station s; otherwise, it is not stored on the edge server 4 of the base station s. The delay t for the user Um to download a file from the small base station 2s m,s = D m,f / ω s log2(1 + γ s,m ), where D m,f is the size of the file f requested by the user Um, ω s is the sub-channel bandwidth of each small base station 2, and γ s,m is the signal-to-interference-plus-noise ratio between the small base station 2s and the user Um. P s is the transmit power of the small base station 2, L0 is a constant, is the distance between the small base station 2s and the user Um, α is the path loss indicator factor, and σ 2 is the thermal noise power spectral density, and P N is the transmit power of the macro base station 3. is the distance between the macro base station 3 and the small base station 2s. The delay t for the small base station 2s to download a file from the macro base station 3N s,N = D m,f / ω N log2(1 + γ N,s ), where ω N is the sub-channel bandwidth of the macro base station 3, and γ N,s is the signal-to-interference-plus-noise ratio between the macro base station 3N and the small base station 2s. The meanings of the parameters in the formula are the same as above. The average delay of all users finally where when the macro base station 3 provides services, k = 1; otherwise, k = 0. In addition, considering the energy consumption limit, the energy consumption for the user Um to download a file from the small base station 2s is E m,s = t m,s×P s The energy consumption of the small base station 2s for downloading a file from the macro base station 3N is E s,N = t s,N ×P s The overall energy consumption of all users is And the total energy consumption of all users should not exceed a certain energy consumption threshold EU. The cache space of each small base station server 4 is CS, and the sum of the sizes of the files initially cached on the edge server 4 of the small base station 2 should not exceed CS, that is The hit rate of user file acquisition is CHf = CH / (CH + CM), which should be greater than the hit rate CHC, where CH is the number of cache hits and CM is the number of cache misses. To sum up, the optimization goal of this problem is to minimize the average delay minT of the users in this system, and the energy consumption, cache space size, and hit rate satisfy the constraint conditions.

[0035] S22 constructs a solution based on the mobile edge caching system and the mathematical model, and uses an improved differential evolution algorithm to solve the base stations that provide services to users.

[0036] In the embodiment of the present invention, the steps of the improved differential evolution algorithm are as follows: Initialize the population NP, dimension D, generation number T, scaling factor F, and crossover probability CR, NP = 200, D = 4, T = 100, F = 0.5, CR = 0.3, and generate the initial population through random numbers randomly generated by the normal distribution in the interval [1, 0]; Evaluate all individuals in the population. For each member in the second subpopulation, the position of the current member is stored in x2. Calculate the result of two vectors: v i = r1·(x2 - x best1 ) - r2·(x2 - x best2 ), where v i is the composite vector, x best1 and x best2 are two best candidate positions, and r1 and r2 are random coefficients. Mutation: Among them, is the target vector, the random exponents r1, r2, and r3 are integers and are not equal to each other, and F is a real constant factor, F > 0. Crossover:

[0037]

[0038] Among them, the value range of rand1ij is [0, 1], representing the jth evaluation of the uniform random number generator, and rand(i) is a randomly selected index. Selection: Respectively substitute the trial vector the current individual vector into the formula to be solved, and judge whether the trial vector can be an individual of the (T + 1)th generation by comparing the sizes of the two values.

[0039] S3 uses collaborative filtering to predict the content that the user is interested in and obtains the predicted content;

[0040] Specific method:

[0041] S31 analyzes and obtains the information of the files downloaded by the user from the user's previous download history to generate a training set and a test set;

[0042] In the embodiment of the present invention, the information of the files downloaded by the user (taking movies as an example) can be analyzed and obtained from the user's previous download history, including movie ID, movie rating, movie type, as well as user ID and timestamp, to generate a training set and a test set.

[0043] S32 trains and tests the collaborative filtering model based on the training set and the test set, and uses the collaborative filtering model to predict the content that the user is interested in.

[0044] In the embodiment of the present invention, the NeuralCF model is trained and tested. Finally, neural network collaborative filtering (NeuralCF) model is used for prediction to obtain the content that the user is interested in according to the prediction result. The MovieLens dataset is used for experiments. This dataset contains approximately 1 million ratings of 3,883 movies by 6,040 users. After obtaining the movie ID, movie rating, movie type, as well as user ID and timestamp of the dataset, the data is filtered according to user preferences and popularity. The user preferences are determined by the ratings of the movies by the user at different times. The popularity of the movies is determined according to the movie ratings. The initial ratings are used to determine the most popular movies for each user, and then the cosine similarity is used to determine the similarity between users, and then the items that these similar users are interested in are recommended. The cosine similarity formula is as follows: The larger this value is, the higher the similarity between the two users, where, are two certain users. Due to the huge amount of data, the method of neural network collaborative filtering (NeuralCF) is adopted to predict the most popular content.

[0045] Take the first 90% of the dataset as the training set and the remaining 10% as the test set. User information (i.e., user ID, longitude / latitude, number of votes, location ID, and score) and movie information (i.e., movie ID, movie rating, movie genre, and timestamp) are input as input data into the NeuralCF model. The CF model consists of multiple layers, where the output of one layer serves as the input to the next layer. Since the present invention uses a pure collaborative filtering model and only uses the user ID and movie ID as input features, and uses encoding to convert them into binary sparse vectors, it is easy to represent users and movies using content features to adjust and solve the cold start problem. The input layer is input into a multi-layer CF architecture, where user and content features are mapped to predicted scores. The final output layer is the predicted score, which is trained by minimizing the pointwise loss between the predicted value and its target.

[0046] S4 caches the predicted content on the base station edge server 4 that serves the user.

[0047] In an embodiment of the present invention, after obtaining the content that the user may be interested in, it is cached on the best server 4 (edge node). When caching, if the cache space is sufficient, it is directly cached; otherwise, the cached file is replaced according to the popularity of the file.

[0048] Please refer to Figure 1 Second, the present invention also provides a mobile edge caching system based on user mobility and preferences, which is applied to the mobile edge caching method based on user mobility and preferences as described in the first aspect above, and is characterized in that;

[0049] It includes a cloud server 1, a small base station 2, a macro base station 3, and a server 4. The cloud server 1, the macro base station 3, the server 4, and the small base station 2 are connected in sequence.

[0050] In an embodiment of the present invention, the server 4 (edge server 4) includes one cloud server 1, N small base stations 2, one macro base station 3, and M users. The user can send a request to the small base station 2. If the data requested by the user exists in the small base station 2, the request of the user is responded to; if the data requested by the user does not exist, the small base station 2 requests data from the macro base station 3 through the server 4. The coverage ranges of each small base station 2 overlap. When the user moves to a certain position, there may be multiple small base stations 2 that can serve the user.

[0051] To better understand the present technical solution, the following embodiments are provided for further illustration:

[0052] Embodiment 1

[0053] Appendix Figure 1It is a diagram of a mobile edge caching system. The system has a three-layer architecture, which from top to bottom are the cloud computing layer, the mobile edge node layer, and the user terminal layer. When the user is moving, the user requests a file from the edge node. When the user is at the overlapping area of the signal coverage of two small base stations 2, both small base stations 2 can provide services for the user. If the file requested by the current user exists in the small base station 2, it is sent to the user. Otherwise, the small base station 2 sends a file request to the macro base station 3, downloads the file from the macro base station 3, and then sends it to the user.

[0054] Embodiment 2

[0055] Appendix Figure 2 It is the flow schematic diagram of the present invention, including the following steps: constructing an edge caching framework, as shown in the appendix Figure 1 shown. Establishing a mathematical model for finding the optimal edge node and constructing a solution: establishing a connectivity availability matrix Cm of the user Um and N small base stations 2, and establishing a matrix Cs of the files stored in the edge server 4 of the base station s. The connectivity availability matrix Cm of the user Um and N small base stations 2 is represented, and where represents the connectivity situation between the user Um and the base station s. If it means that the user Um is not connected to the base station s; if it means that the user Um is connected to the base station s. The total files Ff = {1, 2,..., f,..., F}, the size of each file is Size = {Size1, Size2,..., Sizef,..., SizeF}, and the files stored in the edge server 4 of the base station s are If it means that file 1 is stored on the edge server 4 of the base station s; otherwise, it is not stored on the edge server 4 of the base station s.

[0056] Establishing the delay expression for the user to download a file from the small base station 2 and the delay expression for the small base station 2 to download a file from the macro base station 3: The delay t for the user Um to download a file from the small base station 2s m,s = D m,f / ω s log2(1 + γ s,m ), where D m,f is the size of the file f requested by the user Um, ω s is the sub-channel bandwidth of each small base station 2, γ s,m is the signal-to-dry ratio between the small base station 2s and the user Um, P s is the transmit power of the small base station 2, L0 is a constant, is the distance between the small base station 2s and the user Um, α is the path loss indication factor, σ 2is the thermal noise power spectral density, P N is the transmit power of macro base station 3, is the distance between macro base station 3 and small base station 2s. The delay t for small base station 2s to download a file from macro base station 3N s,N = D m,f / ω N log2(1 + γ N,s ), ω N is the sub-channel bandwidth of macro base station 3, γ N,s is the signal-to-interference-plus-noise ratio between macro base station 3N and small base station 2s, The meanings of the parameters in the formula are the same as above.

[0057] Establish the average delay expression, total energy consumption expression, and file acquisition hit rate expression for all users: The average delay of all users finally where when served by macro base station 3, k = 1, otherwise, k = 0. In addition, considering the energy consumption limit, the energy consumption for user Um to download a file from small base station 2s is E m,s = t m,s × P s , the energy consumption for small base station 2s to download a file from macro base station 3N is E s,N = t s,N × P s , and the total energy consumption of all users is And the total energy consumption of all users should not exceed a certain energy consumption threshold EU.

[0058] Set a reasonable energy consumption threshold. Under the constraints of energy consumption, cache space size, and hit rate, minimize the average delay of users: The cache space of each small base station server 4 is CS, and the sum of the initial cached file sizes on the edge server 4 of small base station 2 should not exceed CS, that is The file acquisition hit rate of users is CHf = CH / (CH + CM), which should be greater than the hit rate CHC, where CH is the number of cache hits and CM is the number of cache misses. To sum up, the optimization goal of this problem is to minimize the average delay minT of the users in this system, and the energy consumption, cache space size, and hit rate satisfy the constraint conditions.

[0059] Based on the edge caching framework and mathematical model, construct a solution, and use the improved differential evolution algorithm to solve for the base station that provides the smallest service delay and lower energy consumption for users.

[0060] Predicting content of user interest: Analyze the information of the files downloaded by the user from the user's previous download history (taking movies as an example), including movie ID, movie rating, movie type, as well as user ID and timestamp, to generate a training set and a test set; secondly, train and test the NeuralCF model; finally, use the Neural Collaborative Filtering (NeuralCF) model for prediction to obtain the content of user interest based on the prediction results.

[0061] Then use NeuralCF to predict the content that the user may be interested in, which is the content that the user may request. Based on this, combined with the optimal node algorithm, calculate the optimal node, and then cache it on the optimal node according to the file popularity. If the cache space is sufficient, cache it directly; otherwise, replace the cached file according to the size of the file popularity. If the edge node providing services to the user caches the file requested by the user, the user directly obtains the file from the edge node; otherwise, the small base station 2 obtains the file from the macro base station 3 to the core network and distributes it to the user.

[0062] Embodiment 3

[0063] Appendix Figure 3 It is a diagram of the NeuralCF model. The CF model consists of multiple layers, including an output layer, a NeuralCF layer, and an input layer. The output of one layer is used as the input of the next layer, and the output layer is the input of the NeuralCF layer, and the NeuralCF layer is the input of the input layer. Since the present invention uses a pure collaborative filtering model, only the user ID and movie ID are used as input features, and they are converted into binary sparse vectors using encoding, and it is easy to represent users and movies using content features to adjust and solve the cold start problem. The input layer is input into a multi-layer CF architecture, where user and content features are mapped to prediction scores. The final output layer is the prediction score, and it is trained by minimizing the pointwise loss between the predicted value and its target.

[0064] What is disclosed above is only a preferred embodiment of a mobile edge caching system and method based on user mobility and preferences of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. A mobile edge caching method based on user mobility and preference, characterized in that: The following steps are involved: Build a mobile edge caching system; The mobile edge cache system searches for a base station that provides services to a user based on the user's movement based on an algorithm; Use collaborative filtering to predict the content that users are interested in and obtain the predicted content; The predicted content is cached on a base station edge server that provides services to users.

2. The mobile edge caching method based on user mobility and preference as claimed in claim 1, It is characterized by: The mobile edge cache system uses an algorithm to find a base station that provides services for the user according to the user's movement: Establish a mathematical model of edge nodes; A solution is constructed based on the mobile edge cache system and the mathematical model, and an improved differential evolution algorithm is used to solve the base station that provides services to users.

3. The mobile edge caching method based on user mobility and preference as claimed in claim 1, It is characterized by: The specific method of using collaborative filtering to predict the content that the user is interested in and obtain the predicted content is as follows: Analyze the user's previous download history to obtain information about the user's downloaded files to generate a training set and a test set; The collaborative filtering model is trained and tested based on the training set and the test set, and the collaborative filtering model is used to predict the content of interest to the user.

4. The mobile edge caching method based on user mobility and preference as claimed in claim 3, characterized in that ； The collaborative filtering model includes an output layer, a NeuralCF layer and an input layer, wherein the output layer is the input of the NeuralCF layer, and the NeuralCF layer is the input of the input layer.

5. A mobile edge caching system based on user mobility and preference, applied to the mobile edge caching method based on user mobility and preference as claimed in claim 1, characterized in that ； It includes a cloud server, a small base station, a macro base station and a server, and the cloud server, the macro base station, the server and the small base station are connected in sequence.

Images