A mobile edge caching method and apparatus based on user location prediction

Abstract

This invention discloses a mobile edge caching method based on user location prediction. For multi-base station caching scenarios, this method caches a portion of files in nodes of a mobile edge network with storage capabilities. When users request these cached files, they can obtain them directly through the edge nodes without sending requests to the server via the backhaul link, avoiding duplicate file transmission and alleviating pressure on the operator's backhaul link. Simultaneously, a user location prediction mechanism is introduced, using an LSTM model to predict user locations, obtain the user group within the base station's service area for a certain time period, and calculate the file request distribution. Under cache capacity constraints, by optimizing the global caching strategy, a problem of maximizing operator caching revenue is constructed, simplified to a knapsack problem, and a dynamic programming algorithm is used to find the optimal solution for the caching strategy of each base station at each time step. This effectively reduces backhaul link overhead and improves the operator's caching revenue.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and specifically to a mobile edge caching method and apparatus based on user location prediction. Background Technology

[0002] The development of IoT and 5G technologies has led to a surge in mobile traffic, significantly increasing the burden on backhaul links. Mobile edge caching technology is considered an effective solution to this problem. This technology caches a portion of files in nodes of a mobile edge network with storage capabilities, allowing users to retrieve these cached files directly from the edge nodes without needing to send requests to the server via the backhaul link. Mobile edge caching avoids duplicate file transfers, alleviates pressure on the operator's backhaul link, reduces latency for users retrieving files, and improves Quality of Service (QoS).

[0003] The key to mobile edge caching technology lies in how to cache files on devices at the network edge. Furthermore, it's necessary to consider the impact of user mobility on caching strategies. Due to user movement, the user groups within the service areas of different base stations are constantly changing, and the distribution of requests for different files by these user groups varies at different times. Since user movement follows certain patterns, user locations can be predicted based on a large amount of user trajectory data, thereby obtaining the user groups within the base station service areas for each time period of the day, and thus determining the request distribution to formulate caching strategies. Summary of the Invention

[0004] Purpose of the Invention: To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a mobile edge caching method based on user location prediction. This method allows base stations to address the problems of excessive backhaul link load and low caching revenue for operators.

[0005] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] Firstly, a mobile edge caching method based on user location prediction is provided, including:

[0007] Based on the user's historical location information, the user's location information at the next moment is predicted using an LSTM-based user location prediction model.

[0008] Based on the predicted user location information for the next moment and combined with the user preference information for the current moment, construct the operator cache revenue maximization problem P1 in the mobile edge scenario;

[0009] The problem of maximizing operator cache revenue P1 is simplified into an optimization subproblem P2;

[0010] The optimal caching strategy of the base station is obtained by using dynamic programming to solve the optimization subproblem P2, and the user preference information is updated according to the user's request in the next time step.

[0011] Based on the optimal caching strategy obtained from the solution, the base station implements mobile edge caching of files.

[0012] The preprocessing described includes missing data imputation, one-hot encoding, and sliding window cutting.

[0013] In some embodiments, the missing data filling includes:

[0014] Suppose that the user's location information is missing at time t, and the nearest time interval with location information is t′, then the method for handling the user's missing location information at time t is as follows:

[0015] (1) |tt′|≤10min, the user's location information at time t is consistent with that at time t′;

[0016] (2)|tt′|>10min, find the location information that appears most frequently at time t in the historical data, and fill the user location data at time t.

[0017] In some embodiments, the one-hot encoding includes: after missing data is filled, performing one-hot encoding on the location and weekday data, wherein the feature dimension of each time step after encoding is the number of locations plus the number of weeks in the user's original dataset.

[0018] In some embodiments, the problem P1 for maximizing operator caching revenue in mobile edge scenarios is constructed, including:

[0019] The base station set is There are a total of F files in the system, and the file set is: There are U users in total, and the user set is...

[0020] The user pays a unit price of v, and the operator incurs a unit cost of b for the backhaul link; a user retrieves a file from access base station j. The operator's revenue falls into two categories:

[0021] (1) The base station j to which the user accesses caches the requested file, and the user downloads the file directly from the base station. The user directly obtains the file and the operator does not need to pay for the backhaul link overhead. The operator's revenue is L. f v;

[0022] (2) If the base station j accessed by the user does not cache the requested file, the operator needs to incur additional backhaul link overhead to retrieve the file from the cloud and then send it to the user. In this case, the operator's revenue is L.f (vb);

[0023] The revenue R of the operator during the time period (t) for:

[0024]

[0025] in, Let f be the caching strategy variable for file f within the region where base station j is located at time t.

[0026] Let t be the average number of times file f is requested within the area where base station j is located. The formula is:

[0027]

[0028] The set of users who appear within the coverage area of ​​base station j at time t is determined by the user location information at the next time moment predicted by the model.

[0029] Let be the preference of user u for file f at time t, i.e., the probability that user u requests file f from the base station; let the initial preferences of all users for all files at time t=0 be... satisfy:

[0030]

[0031] Suppose that each user's initial preferences follow a Zipf distribution, that is, at time t=0, the user's initial preferences are... The probability of requesting file f from the base station for:

[0032]

[0033] Where α is the skewness parameter of the Zipf distribution, and f′ is the variable of f;

[0034] Considering that users typically do not request the same file repeatedly within a day, the user's preference for files already requested is 0 at any given time; the set of files requested by user u before time t is defined as follows. but:

[0035]

[0036] Under the constraint of limited base station buffer capacity, the operator's buffer revenue maximization problem P1 is expressed as:

[0037]

[0038]

[0039]

[0040] Where st represents a constraint, L f The size of file f, constraint C1 refers to the sum of the file sizes cached by each base station. Not exceeding the maximum capacity C of this base station j Constraint C2 refers to the caching strategy variable that needs to be optimized. It is a 0-1 variable.

[0041] In some embodiments, the operator caching revenue maximization problem P1 is simplified into an optimization sub-problem P2, including:

[0042] Users within each service area can only obtain the requested file from the base station in their respective area. Therefore, the caching strategy is determined independently for each base station, meaning the caching strategy variable for each base station in time period t is different. They are uncoupled;

[0043] Therefore, optimization problem P1 is equivalent to J optimization subproblems, where J is the number of base stations;

[0044] Where P2 is the optimization subproblem corresponding to base station j, and its mathematical programming form is as follows:

[0045]

[0046]

[0047]

[0048] in The caching strategy for base station j;

[0049] Further simplification of the objective function of the optimization subproblem P2:

[0050]

[0051] make δ is a constant, and is related to the optimization variables. Irrelevant; furthermore, let the intermediate vector The optimization subproblem P2 is equivalent to:

[0052]

[0053]

[0054]

[0055] The optimal caching strategy for base station j is obtained by solving the optimization subproblem P2 using dynamic programming.

[0056] At time t, the optimal caching strategy C for all base stations is obtained by applying dynamic programming to the optimization subproblems corresponding to these J base stations. t .

[0057] In some embodiments, the optimal caching strategy for the base station is obtained by dynamic programming, including the following steps:

[0058] 1) Based on the file size vector L = [L1, L2, ..., L...] F The buffer capacity C of base station j j User preference distribution at time t and the user group within the service range of base station j Calculate the intermediate vector

[0059] 2) Initialize the dynamic programming matrix P, let P(0,c)=0, c∈{0,1,2,...C} j};

[0060] 3) Let f = 0, execute step 4), iterate F+1 times, and in each iteration f = f+1;

[0061] 4) Let c = 1, execute steps 5)-6), iterate C j Each iteration consists of c = c + 1.

[0062] 5) If the file size L f <c, proceed to step 6); otherwise, let P(f,c) = P(f-1,c), and proceed to step 7);

[0063] 6) Further judgment on V jf +P(f-1,cL f If P(f,c) > P(f-1,c), and this condition is satisfied, let P(f,c) = V. jf +P(f-1,cL f Otherwise, let P(f,c) = P(f-1,c);

[0064] 7) Let the intermediate variable W = C j ;

[0065] 8) Let f = F, execute step 9), iterate F times, with f = f - 1 in each iteration;

[0066] 9) If P(f,W)>P(f-1,W), let W = WL f Otherwise

[0067] 10) Update user preference information for the next time step.

[0068] in The optimal caching strategy for base station j is determined by the solution.

[0069] In a second aspect, the present invention provides a mobile edge caching device based on user location prediction, including a processor and a storage medium;

[0070] The storage medium is used to store instructions;

[0071] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0072] Thirdly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0073] Beneficial effects:

[0074] The significant advantages of this invention compared to existing technologies are:

[0075] 1. The method of the present invention aims to maximize operator revenue. By predicting user location information and combining it with user preferences, it optimizes base station caching decisions, which can significantly improve caching revenue.

[0076] 2. Most studies have not considered the impact of user mobility on caching strategies. Due to user movement, the user groups within the service areas of different base stations are constantly changing, and the distribution of requests for different files by these user groups varies at different times. Since user movement follows certain patterns, this invention proposes a mobile edge caching method based on user location prediction for multi-base station caching scenarios. This method divides a day into several time periods, uses an LSTM model to predict user locations, obtains the user groups within the base station service areas of a given time period, and calculates the file request distribution.

[0077] 3. Under the constraint of cache capacity, an optimization problem maximizing operator benefit was constructed by optimizing the global caching strategy. This problem was simplified into a knapsack problem, and a dynamic programming algorithm was used to find the optimal solution for the caching strategy of each base station. Attached Figure Description

[0078] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention;

[0079] Figure 2 This is a schematic diagram of the system model of the method in an embodiment of the present invention;

[0080] Figure 3 This is a schematic diagram of the LSTM-based user location prediction model in an embodiment of the present invention;

[0081] Figure 4 This is a schematic diagram illustrating the process of generating the training dataset for the prediction model in an embodiment of the present invention.

[0082] Figure 5 , Figure 6 The simulation diagrams for the proposed method in the embodiments are shown under different cache capacities and skewness parameters of the Zipf distribution. Detailed Implementation

[0083] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0084] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0085] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0086] Example 1

[0087] A mobile edge caching method based on user location prediction includes:

[0088] Based on the user's historical location information, the user's location information at the next moment is predicted using an LSTM-based user location prediction model.

[0089] Based on the predicted user location information for the next moment and combined with the user preference information for the current moment, construct the operator cache revenue maximization problem P1 in the mobile edge scenario;

[0090] The problem of maximizing operator cache revenue P1 is simplified into an optimization subproblem P2;

[0091] The optimal caching strategy of the base station is obtained by using dynamic programming to solve the optimization subproblem P2, and the user preference information is updated according to the user's request in the next time step.

[0092] Based on the optimal caching strategy obtained from the solution, the base station implements mobile edge caching of files.

[0093] The mobile edge caching model in this embodiment is as follows: Figure 1 As shown. Consider a multi-base station coverage area, which is divided into J smaller areas, each containing one base station, with the base station set as follows. Divide a day into T time periods, and form a set of time periods. There are U users in the model, and the user set is... This user (U) can choose to stay, move to another smaller area, or leave this service area at any given time. The system contains F files, and the file set is... Each user has an independent preference distribution for these F files within each time period and obtains the files by connecting to a base station in their respective small area. If the base station has pre-cached the files requested by the user, the base station directly sends the files to the user; otherwise, the files are first downloaded from the cloud server to the base station via a backhaul link before being sent to the user, which incurs additional overhead for the operator. This embodiment assumes that the cloud server stores all the files requested by the user.

[0094] definition Let be the preference of user u for file f at time t, i.e., the probability that user u requests file f from the base station. Let the initial preferences of all users for all files at time t=0 be... in, satisfy:

[0095]

[0096] The Zipf distribution is often used to simulate the distribution of click probabilities for different files, i.e., a small number of files are requested a large number of times, while most files are requested very few times. To model user requests, this embodiment assumes that each user's initial preferences follow a Zipf distribution, i.e., at time t=0, the user's initial preferences... The probability of requesting file f is:

[0097]

[0098] Here, α is the skewness parameter of the Zipf distribution; the larger the value, the more concentrated the request distribution. Furthermore, the user's preference distribution is different at each time point. This is because users typically do not repeatedly request the same file within a day; therefore, at each time point, the user's preference for files already requested is 0. Let the set of files requested by user u before time t be defined as... but:

[0099]

[0100] (1) Modeling the optimization problem

[0101] At the end of each time period, the operator re-determines the caching strategy for the base stations in the next time period based on user group predictions. Considering the limited cache capacity of each base station, the following applies:

[0102]

[0103] Among them, C j Let L be the buffer capacity of base station j. f The size of file f, Let be the caching strategy variable for base station j at time t. like This indicates that file f is cached in base station j at time t. The caching strategy for all base stations at time t is C. t , It is an F×J dimensional matrix consisting of 0-1 variables.

[0104] Each time a user requests a file, they pay the operator based on the file size. If the requested file is pre-cached in the base station, the user can directly obtain the file, and the operator does not need to pay backhaul link costs. If the file is not pre-cached in the base station, the operator needs to incur additional backhaul link costs to retrieve the file from the cloud before sending it to the user. The user's unit payment is v, and the operator's unit backhaul link cost is b. Based on the above analysis, the operator's revenue during time period t is:

[0105]

[0106] in, Let t be the average number of times file f is requested within the area where base station j is located. The formula for calculating this number is:

[0107]

[0108] The goal of this embodiment is to increase the operator's revenue. Under the constraint of limited base station cache capacity, the problem of maximizing operator cache revenue can be expressed as:

[0109]

[0110]

[0111]

[0112] C1 limits the sum of the file sizes cached by each base station to no more than its maximum capacity, while C2 limits the cache policy variables that need to be optimized. It is a 0-1 variable.

[0113] (2) Dataset preprocessing and prediction model training

[0114] Considering that user locations change constantly, meaning the user population within the service area of ​​each base station varies across different time periods, the base station needs to determine its caching strategy for the next time period at the end of each time period. Furthermore, user movement trajectories follow certain patterns. Predicting user movement locations and using this predicted location information to calculate user request distribution can help formulate future base station caching strategies. LSTM neural networks, which have repetitive modules and can learn long-term dependencies, are used in this invention to predict user locations. The prediction model is as follows: Figure 2 As shown. The dataset includes the arrival time and dwell time of each user at different labeled locations. To facilitate model training, the user's location at each full hour is extracted as valid data. Considering that the user's dataset may not be complete, and location information at some full hours may be missing, while the training of the LSTM model requires continuous temporal data, it is necessary to fill in the missing data in the dataset. Assuming that the user's location information at time t is missing, and the nearest time interval with location information is t′, then the method for handling the missing location information of the user at time t is as follows:

[0115] (3) |tt′|≤10min, the user's location information at time t is consistent with that at time t′;

[0116] (4) |tt′|>10min, find the location information that appears most frequently at time t in the historical data, and fill in the user location data at time t. For example, if the user appears at location A most often at 10:00 am on Monday in the historical data, but the location data for 10:00 am on a certain Monday is missing, then use location A to fill in the location information at that time.

[0117] After missing data imputation, this embodiment performs one-hot encoding on the location and weekday data, using it as input data for the user location prediction model. This embodiment's LSTM-based user location prediction model employs multi-dimensional single-step prediction, predicting the next step's feature (location) by inputting the features (location, weekday) from the previous N steps. Specifically, the one-hot encoded data is sliced ​​using a sliding window to obtain training and test sets. The feature dimension for each step is the sum of the number of locations and the number of weeks in the user's original dataset, such as... Figure 3 As shown. For example, a user's dataset has a length of n and has reached L positions. At a certain time on a Thursday of a certain week, the user appears at position l. Then, the feature vector of this user at that time divided by k... l ,k L+4 Except for 1, all others are 0.

[0118] LSTM training consists of three phases: a forgetting phase, a selective memory phase, and an output phase, corresponding to the forget gate, input gate, and output gate in its structure, respectively. These gates determine whether to retain or discard information from the input data. During training, these gates can pass relevant information (memory) down the training process, thereby enabling prediction.

[0119] By training the LSTM model, the user's location at the next moment can be predicted, i.e., within the coverage area of ​​which base station the user will appear in. In this embodiment, Let be the set of users who appear within the coverage area of ​​base station j at time t. This set is obtained through a user location prediction model.

[0120] (3) Further simplification of the optimization problem

[0121] Analyzing the operator's caching revenue maximization problem P1 reveals that users within each service area can only obtain requested files from base stations in their respective areas. Therefore, the caching strategy for each base station is independent, meaning the caching strategy variable for each base station during time period t is... They are uncoupled. Therefore, optimization problem P1 can be equivalent to J optimization subproblems, where J is the number of base stations. Specifically, P2 is the optimization subproblem corresponding to base station j, and its mathematical programming form is as follows:

[0122]

[0123]

[0124]

[0125] The objective function of the optimization subproblem P2 is further simplified below:

[0126]

[0127] make δ is a constant, and is related to the optimization variables. It is irrelevant. Furthermore, let At this point, the optimization subproblem P2 can be equivalent to:

[0128]

[0129]

[0130]

[0131] This reveals that it is a standard 0-1 single knapsack problem, meaning that solving the optimization subproblem P2 is equivalent to solving a 0-1 single knapsack problem. Dynamic programming is a good algorithm for solving knapsack problems. Therefore, this embodiment uses dynamic programming to solve the optimization subproblem P2 to obtain the optimal caching strategy for each base station.

[0132] (4) Solve the optimization problem.

[0133] At time t, this embodiment solves the optimization subproblems corresponding to these J base stations using dynamic programming algorithms to obtain the optimal caching strategy C for all base stations. t The core idea of ​​dynamic programming is that if the optimal solution to a problem can be derived from the optimal solutions to its subproblems, then the optimal solutions to the subproblems can be solved first, and then the optimal solution to the original problem can be constructed. Furthermore, these subproblems can utilize the solutions to other subproblems to obtain their own solutions, which reduces the algorithm's complexity. When using dynamic programming to solve the 0-1 single knapsack problem, the relationship between the original problem and a set of smaller knapsack problems can be identified, and the solution to the original problem can be obtained by solving the solutions to these smaller knapsack problems. Based on the above analysis, a matrix P(0...F,0...C) is constructed at time t. j The matrix is ​​used to solve the optimization subproblem corresponding to base station j. The purpose of this matrix is ​​to store the maximum reward corresponding to the optimal solution of these smaller knapsack problems. The value is: when the buffer capacity of base station j is c, considering only the first f files, the maximum revenue of base station j is calculated as follows:

[0134]

[0135]

[0136]

[0137] Furthermore, given that the base station's cache capacity is c, and the optimal caching strategy for the first f-1 files is obtained, and a decision is made on whether to cache the f-th file, the following cases are discussed:

[0138] 1) If the base station's cache capacity is less than the size of the f-th file, i.e., L f If the value is greater than c, then the base station cannot cache file f. The caching strategy is the same as the optimal caching strategy that only considers the first f-1 files, and the maximum benefit is also the same as the maximum benefit of the first f-1 files, i.e., P(f,c) = P(f-1,c).

[0139] 2) If the base station's cache capacity is greater than or equal to the size of the f-th file, i.e., L fIf f ≤ c, then we need to compare the benefits of caching and not caching this file to decide whether to cache it. If we don't cache file f, the maximum benefit at this point is the same as the maximum benefit considering the first f files, which is P(f-1, c); if we cache file f, then the maximum benefit at this point is the benefit V of caching file f. jf The maximum profit P(f-1, cL) corresponding to the remaining cache capacity after adding cached file f. f Compare the returns of the two options and select the larger value as the value of P(f,c), i.e., P(f,c) = max{P(f-1,c),V} jf +P(f-1,cL f )}.

[0140] Based on the above analysis, we can derive the recursive formula for maximizing the profit of these knapsack problems:

[0141]

[0142] Next, the optimal caching strategy for base station j is obtained using the obtained matrix P. If the maximum caching benefit P(f,c) for the first f files is the same as the maximum caching benefit P(f-1,c) for the first f-1 files, then file f is not cached; otherwise, file f is cached in base station j. Therefore, a dynamic programming-based caching strategy optimization method can be derived, comprising the following steps:

[0143] 1) Based on the file size vector L = [L1, L2, ..., L...] F The buffer capacity C of base station j j User preference distribution at time t and the user group within the service range of base station j Calculate the vector

[0144] 2) Initialize matrix P, let P(0,c)=0, c∈{0,1,2,...C} j}

[0145] 3) Let f = 0, execute step 4), iterate F+1 times, with f = f+1 in each iteration.

[0146] 4) Let c = 1, execute steps 5)-6), iterate C j Each iteration is c = c + 1.

[0147] 5) If L f <c, proceed to step 6), otherwise let P(f,c) = P(f-1,c), proceed to step 7).

[0148] 6) Further judgment on V jf+P(f-1,cL f If P(f,c) > P(f-1,c), and this condition is satisfied, let P(f,c) = V. jf +P(f-1,cL f Otherwise, let P(f,c) = P(f-1,c).

[0149] 7) Let W = C j

[0150] 8) Let f = F, execute step 9), iterate F times, with f = f-1 in each iteration.

[0151] 9) If P(f,W)>P(f-1,W), let W = WL f Otherwise

[0152] 10) Update user preference information for the next time step.

[0153] In summary, this embodiment proposes a mobile edge caching method based on user location prediction for mobile edge caching scenarios. Using caching strategy as the optimization variable and constrained by cache capacity, an optimization problem maximizing operator caching revenue is constructed and simplified into a 0-1 knapsack problem. This problem is then solved using a dynamic programming algorithm incorporating user location prediction information.

[0154] Figure 5 The impact of base station cache capacity and number of users on operator revenue is presented in the figure. As can be seen from the figure, with a fixed cache capacity, operator revenue increases with the increase in the number of users. This is because an increase in the number of users leads to an increase in file clicks, meaning more users will pay, thus increasing operator revenue. Furthermore, operator revenue also increases with the increase in cache capacity per base station, because a larger cache capacity can cache more files, saving more backhaul link overhead. The increase in operator revenue from increasing cache capacity is even more significant when the number of users increases.

[0155] Figure 6 The impact of changing the Zipf distribution parameter α on operator revenue is presented. A larger α results in greater operator revenue, while increasing cache capacity has a smaller effect on increasing operator revenue. This is because a larger α leads to a more concentrated distribution of user preferences, creating a head effect, thus requiring only a small amount of cache capacity to generate sufficient revenue.

[0156] Example 2

[0157] Secondly, this embodiment provides a mobile edge caching device based on user location prediction, including a processor and a storage medium;

[0158] The storage medium is used to store instructions;

[0159] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0160] Example 3

[0161] Thirdly, this embodiment provides a storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0162] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0164] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0165] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0166] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0167] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A mobile edge caching method based on user location prediction, characterized in that, Comprise: According to the user historical location information, the user location information at the next time is predicted by using the user location prediction model based on LSTM; According to the predicted user location information at the next time, the user preference information at the current time is combined to construct the operator cache revenue maximization problem P1 in the mobile edge scenario, specifically comprising: The set of base stations is There are F files in the system, and the set of files is There are U users, and the set of users is The unit price paid by the user is v, and the unit cost of the backhaul link required by the operator is b; a user obtains a file from the access base station j The operator's revenue is divided into two cases: (1) the base station j accessed by the user caches the requested file, and the user directly downloads the file from the accessed base station; the user directly obtains the file and the operator does not need to pay the backhaul link cost, and the operator's revenue is L f v; (2) the base station j accessed by the user does not cache the requested file, and the operator needs to pay additional backhaul link cost to obtain the file through the cloud and then send it to the user, at which time the operator's revenue is L f (v-b); The operator receives revenue R during the specified time period. (t) for: wherein, is the cache policy variable for file f in the area where base station j is located at time t. For t time, the average number of times that file f is requested in the area where base station j is located, the calculation formula is: For the set of users present in the coverage area of base station j at time t, the user location information at the next time instant is determined by the model prediction. is the preference of user u for file f at time t, i.e., the probability that user u requests file f from the base station; and the initial preference of all users for all files at time t = 0 is satisfies: Let the initial preference of each user obey Zipf distribution, i.e. at t = 0, the user Probability of requesting file f from the base station is: Wherein α is the skewness parameter of Zipf distribution, f' is the variable of f; Considering that a user usually does not request the same file within a day, the preference of a user for a file that has been requested before is defined as 0 at each time instant; the set of files requested by user u before time t is defined as then: Under the constraint of limited base station cache capacity, the operator cache revenue maximization problem P1 is expressed as: where s.t. denotes constraints, L f The size of file f, constraint C1 refers to the sum of the size of files cached by each base station Not more than the maximum capacity C of this base station j Constraint C2 refers to the cache strategy variable that needs to be optimized is a 0-1 variable; The operator cache revenue maximization problem P1 is simplified to the optimization subproblem P2; The optimal cache strategy of the base station is obtained by solving the optimization subproblem P2, and the user preference information at the next time is updated according to the user's request; According to the optimal cache strategy of the base station obtained, the mobile edge cache of the file is implemented by the base station. 2.The mobile edge caching method based on user location prediction according to claim 1, wherein, According to the user historical location information, the user location information at the next time is predicted by using the user location prediction model based on LSTM, comprising: Obtain the user historical location information with time label as data set; Data preprocessing is performed on the data set to obtain training set and test set; The training set and test set are used to train and optimize the user location prediction model based on LSTM to obtain the trained user location prediction model based on LSTM; The location information of the previous N times is input into the trained user location prediction model based on LSTM, and the predicted user location information at the next time is obtained according to the output of the model. 3.The mobile edge caching method based on user location prediction according to claim 2, characterized in that, The preprocessing includes missing data filling, one-hot coding and sliding window cutting. 4.The mobile edge caching method based on user location prediction according to claim 3, wherein, The missing data filling includes: Suppose the user location information at time t is missing, and the time interval of the time with location information closest to t is t', then the processing method of the user location information missing at time t is as follows: |t-t'|≤10min, the user location information at time t is consistent with that at time t'; |t-t'|>10min, find the location information with the maximum frequency at time t in the historical data, and fill the user location data at time t. 5.The mobile edge caching method based on user location prediction according to claim 3, wherein, The one-hot coding includes: after the missing data filling is completed, the position and the week data are one-hot coded, and the feature dimension of each time after coding is the position number plus the week number in the original data set of the user. 6.The mobile edge caching method based on user location prediction according to claim 1, wherein, The operator cache revenue maximization problem P1 is simplified to the optimization subproblem P2, comprising: The users in each service area can only obtain the requested files from the base stations in the area, so the formulation of the cache strategy for each base station is independent of each other, that is, the cache strategy variable of each base station at time t are uncoupled; Therefore, the optimization problem P1 is equivalent to J optimization subproblems, and J is the number of base stations; Wherein P2 is the optimization subproblem corresponding to base station j, and the mathematical programming form is as follows: wherein is the buffer policy of base station j; The objective function of the optimization subproblem P2 is further simplified: Let δ be a constant, independent of the solution of the optimization variable ; furthermore, let the intermediate vector be defined as P2 is equivalent to: The optimal caching strategy of base station j is obtained by solving the optimization sub-problem P2 using dynamic programming At time t, the optimal caching strategy C of all base stations is obtained by using dynamic programming for the optimization sub-problems corresponding to the J base stations respectively t . 7.The mobile edge caching method based on user location prediction according to claim 1, wherein, The optimal cache strategy of the base station is obtained by using dynamic programming, comprising the following steps: 1) According to the file size vector L = [L1, L2,..., L F ], the base station j's buffer capacity C j , the user preference distribution at time t, and the user population within the base station j's service range, the intermediate vector 2) initialize the dynamic programming matrix P, let P(0,c) = 0, c e {0,1,2,...C j}; 3) let f=0, execute step 4), iterate F+1 times, f=f+1 each time; 4) Let c = 1, perform steps 5) - 6), iterate C j each time c = c + 1; 5) if file size L f <c, execute step 6), otherwise let P(f,c) = P(f-1,c), execute step 7); 6) further judge V jf + P(f-1,c-L f ) > P(f-1,c), if satisfied, let P(f,c) = V jf + P(f-1,c-L f ), otherwise let P(f,c) = P(f-1,c); 7) Let the intermediate variable W = C j ; 8) let f=F, execute step 9), iterate F times, f=f-1 each time; 9) if P(f, W) > P(f-1, W), let W = W - L f , else let 10) updating the user preference information for the next time 8.A mobile edge caching apparatus based on user location prediction, characterized in that, Comprise processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 7.

9. A storage medium having stored thereon a computer program, characterized in that The computer program, which is executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images