Data caching method based on popular content prediction in edge computing
By adopting a data caching method based on popular content prediction in edge computing, using ARIMA model to predict content popularity, and using dynamic programming and approximation algorithms, the challenges of data caching strategies in edge computing are solved, and the user experience is guaranteed and service is achieved. Maximize provider revenue.
Patent Information
- Application Number
- CN202211199842.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-29
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-09-29
AI Technical Summary
In the edge computing scenario, how to design a data caching strategy to ensure user experience and maximize the benefits of service providers, considering the capacity limitations and specifications of edge servers.
The data cache method based on popular content prediction is adopted, and a unified formal data cache problem is constructed, and the content popularity is predicted using differential integrated moving average autoregression model ARIMA, and dynamic programming algorithms and approximation algorithms are used to solve the large-scale data cache problem.
It achieves the maximization of service provider benefits while ensuring user experience, and determines the data sets that need to be cached by accurately predicting content popularity, improving the efficiency and accuracy of data caching.
Smart Images

Figure CN115633091B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a data caching method based on popular content prediction in edge computing, and belongs to the field of data caching. Background Art
[0002] Since there will be a large communication delay when users send requests to cloud data centers, caching data and content on edge nodes closer to users can not only reduce communication overhead, but also reduce the pressure on the backhaul network. Especially for data such as videos, images and files, content caching can significantly reduce service delays. By storing content in geographically dispersed caches and using the load balancing, content distribution and scheduling functional modules of the central platform, users can obtain the required content nearby and reduce network congestion.
[0003] In edge computing, data caching issues need to be considered, as well as the capacity limitations and specification characteristics of edge servers. What kind of data caching strategy should be adopted in edge computing scenarios to ensure user experience and maximize the benefits of service providers is a challenging and practical research issue.
[0004] In view of this, it is necessary to propose a new data caching method based on popular content prediction in edge computing to solve the above problems. Summary of the invention
[0005] The object of the present invention is to provide a data caching method based on popular content prediction in edge computing to ensure user experience while maximizing the revenue of the service provider.
[0006] To achieve the above object, the present invention provides a data caching method based on popular content prediction in edge computing, comprising:
[0007] Step 1: Input the historical data of a certain area, perform data preprocessing, and determine whether the historical data time series is a difference stationary series. If so, construct the difference integrated moving average autoregressive model ARIMA and proceed to step 2;
[0008] Step 2: Construct a unified formal data caching problem:
[0009]
[0010]
[0011]
[0012]
[0013] Where V = {v1,v2,…,v n} represents a set of edge servers in a certain area, C = {c1, c2, ..., c n} represents the resource capacity of each edge server, D={d1,…,d l} is a set of preprocessed data determined in step 1, S = {s1, s2, …, s l} represents the size of the data, U={u1,u2,…,u m} represents the user set in this area, a certain data d t Cache at edge server v j The revenue is expressed as p t express, Indicates whether user i goes to edge server j to request data d t , if user i goes to edge server j to request data d t ,but The value is 1, otherwise it is 0. Indicates data d t Is it cached on edge server j? If data d is cached on edge server j, t ,but The value of is 1, otherwise it is 0, defining the unit cost of cache data on the edge server as cost;
[0014] Step 3: Initialize the edge server coverage H, user set U, edge server set V, and select an edge server v from the edge server set V j , let the resource capacity of the server be c j ;
[0015] Step 4: If the edge server v j User i in the coverage area has requested data d t demand, that is When the value of is 1, p t ←p t -s t cost, otherwise p t ←0, repeat step 4 until all data are traversed;
[0016] Step 5: Use FPTAS to analyze the data d t The revenue p t After preprocessing, we get p t ′ is the value of the data benefit actually used in the subsequent algorithm;
[0017] Step 6: Use the dynamic programming algorithm to solve the formal data caching problem in step 2. t ' as data d tThe benefit of caching to edge server j is to create a two-dimensional array B[t][m], which represents the highest benefit of loading the edge server with capacity m in the first t data;
[0018] Step 7: Initialize B[t][m] and perform dynamic programming to solve it. The recursive formula of dynamic programming is defined as:
[0019]
[0020] Step 8: After the dynamic programming solution is completed, pruning operation is performed. If the data d t It has been cached on edge server j, that is, The value of is 1. At the same time, if there is user i within the coverage area of server j, then Ensure that in the next iteration, there is no need to process user i's data d t Requests;
[0021] Step 9: Repeat steps 3 to 8 until the server has no remaining capacity or no other data needs to be cached, and then output the data cache solution S * And the corresponding data cache benefit R.
[0022] As a further improvement of the present invention, the data preprocessing is specifically to perform randomness and stationarity detection on the time series of historical data, so as to determine the parameters for constructing the differential autoregressive moving average model ARIMA.
[0023] As a further improvement of the present invention, the ARIMA model is written as ARIMA (p, d, q), which includes three parameters, where p is the autoregressive order, d is the number of differences, and q is the moving average order.
[0024] As a further improvement of the present invention, the randomness and stability detection is specifically as follows:
[0025] Step 11: Perform a stationarity test on the time series of historical data. If the p-value corresponding to the unit root test statistic is greater than 0.05, it is a non-stationary series and proceed to step 12;
[0026] Step 12: Perform differential operation on the historical data time series until the time series becomes a stationary series;
[0027] Step 13: Perform white noise test. If it is a white noise sequence, end directly. Otherwise, it means that the time series is a non-white noise sequence, and go to step 14;
[0028] Step 14: Determine the model order of the processed stationary non-white noise sequence;
[0029] Step 15: After the model is determined, the residual sequence needs to be tested for white noise. If the residual sequence is a white noise sequence, it means that there is no useful information in the residual, and the detection process ends. If it is a non-white noise sequence, repeat step 14.
[0030] As a further improvement of the present invention, the differential operation in step 12 performs a second-order differential at most.
[0031] As a further improvement of the present invention, step 13 performs a white noise test using the LB statistic, and then calculates the corresponding p value. If the p value is significantly greater than the significance level of 0.05, it is a white noise sequence.
[0032] As a further improvement of the present invention, in step 14, the maximum likelihood ratio method is used to estimate the parameters of the model, estimate the values of various parameters, and then for each different model, the BIC information criterion is used to determine the model order and determine the p and q parameters.
[0033] As a further improvement of the present invention, in step 3, the formalized data caching problem constructed in step 2 is regarded as a dynamic multi-dimensional knapsack problem and solved using the MKP algorithm.
[0034] As a further improvement of the present invention, in step 5, FPTAS is used to analyze the above data d t The revenue p t The preprocessing is as follows:
[0035] Step 51: Take the maximum value of the above data benefits and set it as P, that is, P←max{p t |1≤t≤D};
[0036] Step 52: Let θ∈(0,1], let For all data d t ,make
[0037] As a further improvement of the present invention, the optimization problem after step 5 has been transformed into a 0-1 knapsack problem, which is processed by a dynamic programming algorithm in step 6.
[0038] The beneficial effects of the present invention are as follows: the data caching method based on popular content prediction in edge computing of the present invention constructs a unified formalized data caching problem, formalizes the data caching problem under edge computing, considers the specification differences, capacity limitations and caching problems of multiple data of edge servers, and designs the system objective function to maximize the total revenue of the service provider caching data, thereby maximizing the revenue of the service provider while ensuring user experience; and adopts the differential integrated moving average autoregressive model ARIMA to predict the content popularity of a certain area and determine the data set that needs to be cached, which is more accurate; secondly, for the constructed unified formalized data caching problem, a data caching approximation algorithm is proposed to solve the data caching problem in large-scale scenarios, and the algorithm has a good approximation ratio. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a step diagram of the present invention.
[0040] Figure 2 It is a schematic diagram of a data cache system of the present invention. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0042] like Figure 1 and Figure 2 As shown, the present invention discloses a data caching method based on popular content prediction in edge computing, comprising:
[0043] Step 1: Input the historical data of a certain area, perform data preprocessing, and determine whether the historical data time series is a difference stationary series. If so, construct the difference integrated moving average autoregressive model ARIMA and proceed to step 2;
[0044] Step 2: Construct a unified formal data caching problem:
[0045]
[0046]
[0047]
[0048]
[0049] Where V = {v1,v2,…,v n} represents a set of edge servers in a certain area, C = {c1, c2, ..., c n} represents the resource capacity of each edge server, D={d1,…,d l} is a set of preprocessed data determined in step 1, S = {s1, s2, …, s l} represents the size of the data, U={u1,u2,…,u m} represents the user set in this area, a certain data d t Cache at edge server v j The revenue is expressed as p t express, Indicates whether user i goes to edge server j to request data d t , if user i goes to edge server j to request data d t ,but The value is 1, otherwise it is 0. Indicates data d t Is it cached on edge server j? If data d is cached on edge server j, t ,but The value of is 1, otherwise it is 0, defining the unit cost of cache data on the edge server as cost;
[0050] Step 3: Initialize the edge server coverage H, user set U, edge server set V, and select an edge server v from the edge server set V j , let the resource capacity of the server be c j ;
[0051] Step 4: If the edge server v j User i in the coverage area has requested data d t demand, that is When the value of is 1, p t ←p t -s t cost, otherwise p t ←0, repeat step 4 until all data are traversed;
[0052] Step 5: Use FPTAS to analyze the data d t The revenue p t After preprocessing, we get p t ′ is the value of the data benefit actually used in the subsequent algorithm;
[0053] Step 6: Use the dynamic programming algorithm to solve the above problem. Step 5 t ' as data d t The benefit of caching to edge server j is to create a two-dimensional array B[t][m], which represents the highest benefit of loading the edge server with capacity m in the first t data;
[0054] Step 7: Initialize B[t][m], and the recursive formula of dynamic programming is defined as:
[0055]
[0056] Step 8: After the dynamic programming solution is completed, pruning operation is performed. If the data d t It has been cached on edge server j, that is, The value of is 1. At the same time, if there is user i within the coverage area of server j, then Ensure that in the next iteration, there is no need to process user i's data d t Requests;
[0057] Step 9: Repeat steps 3 to 8 until the server has no remaining capacity or no other data needs to be cached, and then output the data cache solution S * And the corresponding data cache benefit R.
[0058] There will be a large communication delay when users send requests to cloud data centers. Caching data and content on edge nodes closer to users can not only reduce communication overhead, but also reduce the pressure on the backhaul network. Especially for data such as videos, images and files, content caching can significantly reduce service delays. By storing content in geographically dispersed caches and using the load balancing, content distribution and scheduling functional modules of the central platform, users can obtain the required content nearby and reduce network congestion.
[0059] The purpose of the present invention is to disclose a data caching method based on popular content prediction in edge computing, which is used to predict the data caching situation predicted in popular content.
[0060] First, in step 1, historical data of a certain area is input, and data preprocessing is performed on the historical data to determine a popular data set for the area.
[0061] The prediction models for historical data include the following:
[0062] (1) Autoregressive model AR: If the value of a time series at any time can be expressed as the following regression equation:
[0063]
[0064] The time series is said to obey a p-order autoregressive process, denoted as AR(p).
[0065] In the above formula, x t ,x t-1 ,x t-p is the value of the series at different times, is the autoregression coefficient; u t is a white noise term. The AR model uses the autocorrelation of the time series to construct a regression equation, so it is called an autoregressive process.
[0066] (2) Moving average model MA, if a time series can be expressed as the following regression equation at any time:
[0067] x t =u t +θ1u t-1 +θ2u t-2 +…+θ q u t-q
[0068] The time series is said to obey the q-order moving average process, denoted as MA(q).
[0069] In the above formula, x t is the value of the time series at time t, θ1, θ2, …, θ q is the moving regression equation coefficient, u t-1 ,u t-2 ,…,u t-q is the white noise term of different time series. The MA model assumes that the value of the time series at a certain moment is equal to the weighted sum of white noise, which can be used as a supplement to the autoregressive process.
[0070] (3) Regression Moving Average Model ARMA. The ARMA model is a combination of the autoregressive model and the moving average model, so it contains two orders and can be expressed as ARMA (p, q). Among them, p is the autoregressive order, q is the moving average order, and the regression equation of ARMA can be expressed as:
[0071]
[0072] The ARMA model combines the advantages of the AR model and the MA model, and is more effective and commonly used. In the ARMA model, the autoregressive process is responsible for quantifying the relationship between the current data and the previous data, and the moving average process is responsible for solving the problem of white noise.
[0073] (4) Autoregressive Moving Average (ARIMA) Model. AR model, MA model and ARMA model are only applicable to the analysis of stationary time series. If the time series has a long-term upward or downward trend, the ARIMA model needs to be used.
[0074] In real life, many time series are non-stationary, and their mean and variance change over time. However, non-stationary time series are generally stationary time series after difference (subtraction of indicator values at adjacent time points). Therefore, the ARIMA model is written as ARIMA (p, d, q), which includes 3 parameters: p is the autoregressive order, d is the number of differences, and q is the moving average order.
[0075] In step 1, data preprocessing is to perform randomness and stationarity tests on the time series of historical data. According to the test results, the series can be divided into different types, and different analysis methods will be adopted for different types of series.
[0076] For a pure random sequence, also known as a white noise sequence, there is no correlation between the items in the sequence, and the sequence is undergoing completely disordered random fluctuations, and the analysis of the sequence can be terminated. A white noise sequence is a stationary sequence from which no information can be extracted.
[0077] For a stationary non-white noise sequence, its mean and variance are constants. For a non-stationary sequence, since its mean and variance are unstable, the processing method is generally to transform it into a stationary sequence, so that the analysis method of stationary time series can be applied and the ARMA model can be established to conduct corresponding research. If a time series is stationary after differential operation, then the series is a differential stationary series and can be analyzed using the ARIMA model.
[0078] The present application determines whether the historical data time series is a differential stationary series. If so, a differential integrated moving average autoregressive model ARIMA is constructed and the subsequent steps are performed to determine the parameters for constructing the differential autoregressive moving average model ARIMA.
[0079] The randomness and stability tests are as follows:
[0080] Step 11: Perform a stationarity test on the time series of historical data. If the p-value corresponding to the unit root test statistic is greater than 0.05, it is a non-stationary series and proceed to step 12.
[0081] Step 12: Perform differential operation on the historical data time series until the time series becomes a stationary series;
[0082] Step 13: Perform white noise test. If it is a white noise sequence, end directly. Otherwise, it means that the time series is a non-white noise sequence, and go to step 14;
[0083] Step 14: Determine the model order of the processed stationary non-white noise sequence;
[0084] Step 15: After the model is determined, the residual sequence needs to be tested for white noise. If the residual sequence is a white noise sequence, it means that there is no useful information in the residual, and the detection process ends. If it is a non-white noise sequence, repeat step 14.
[0085] In step 11, there are two methods for testing the stationarity of the sequence. One is a graphical test that makes judgments based on the characteristics of the time series graph and the autocorrelation graph. This method is simple to operate and widely used, but its disadvantage is that it is subjective. The other is a method of constructing a test statistic for testing. The present invention uses a unit root test.
[0086] When it is confirmed as a non-stationary sequence in step 11, step 12 is entered to perform a differential operation on the original data until the time series becomes a stationary sequence. Generally, at most a second-order difference is performed here.
[0087] After passing the stationarity test in steps 11 and 12, proceed to step 13 to perform a white noise test. Step 13 performs the white noise test using the LB statistic, and then calculates the corresponding p-value. If the p-value is significantly greater than the significance level of 0.05, it is a white noise sequence, and the analysis of the sequence is terminated.
[0088] If it is a non-white noise sequence, go to step 14 and use the maximum likelihood ratio method to estimate the parameters of the model, estimate the values of each parameter, and then use the BIC information criterion to determine the model order for each different model and determine the p and q parameters.
[0089] After the model is determined, it is necessary to perform a white noise test on the residual sequence. If the residual sequence is a white noise sequence, it means that there is no useful information in the residual. At this time, the determined ARIMA (p, d, q) model can be used for prediction.
[0090] The optimization goal of the formal data caching problem constructed in step 2 is to maximize the benefits that service providers gain by caching data on edge servers to improve user experience.
[0091] In step 3, the formalized data caching problem constructed in step 2 is regarded as a dynamic multidimensional knapsack problem and solved using the MKP algorithm. Therefore, it is natural to adopt a greedy strategy to determine the caching scheme of one edge server at a time.
[0092] In one embodiment of the present application, the edge server v j The capacity limit is c j =67, assuming that the relevant information of the data set D is as shown in Table 1.
[0093]
[0094]
[0095] Table 1: Data information table
[0096] Step 4 traverses all data and confirms data d tThe profit p stored in different edge servers t .
[0097] In step 5, FPTAS is used to analyze the above data d t The revenue p t The specific preprocessing is as follows:
[0098] Step 51: Take the maximum value of the above data benefits and set it as P, that is, P←max{p t |1≤t≤D};
[0099] Step 52: Let θ∈(0,1], let For all data d t ,make
[0100] After step 5, the optimization problem has been transformed into a 0-1 knapsack problem. t ' as data d t The revenue of caching to edge server j is therefore processed by a dynamic programming algorithm in step 6.
[0101] After the dynamic planning solution is completed, the edge server v is obtained j The data set D of the cache strategy on the d = {d1, d4, d8}. Perform pruning operation. If the data d t Already cached on edge server v j On, that is The value of is 1. At the same time, the edge server v j If there is user i in the coverage area, This ensures that in the next iteration, there is no need to process user i's data d t Request.
[0102] Repeat steps 3 to 8 until the server has no remaining capacity or no other data needs to be cached, and then output the data cache solution S * And the corresponding data cache benefit R.
[0103] In summary, the data caching method based on popular content prediction in edge computing of the present invention constructs a unified formalized data caching problem, formalizes the data caching problem under edge computing, considers the specification differences, capacity limitations and caching problems of multiple data of edge servers, and designs the system objective function to maximize the total revenue of the service provider caching data, thereby maximizing the revenue of the service provider while ensuring the user experience; and adopts the differential integrated moving average autoregressive model ARIMA to predict the content popularity of a certain area and determine the data set that needs to be cached, which is more accurate; secondly, for the unified formalized data caching problem constructed, a data caching approximation algorithm is proposed to solve the data caching problem in large-scale scenarios, and the algorithm has a good approximation ratio.
[0104] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A data caching method based on popular content prediction in edge computing, characterized in that: include Step 1: Input the historical data of a certain area, perform data preprocessing, and determine whether the historical data time series is a difference stationary series. If so, construct the difference integrated moving average autoregressive model ARIMA and proceed to step 2; Step 2: Construct a unified formal data caching problem: Where V = {v1,v2,…,v n } represents a set of edge servers in a certain area, C = {c1, c2, ..., c n } represents the resource capacity of each edge server, D={d1,…,d l } is a set of preprocessed data determined in step 1, S = {s1, s2, …, s l } represents the size of the data, U={u1,u2,…,u m } represents the user set in this area, a certain data d t Cache at edge server v j The revenue is expressed as p t express, Indicates whether user i goes to edge server j to request data d t , if user i goes to edge server j to request data d t ,but The value is 1, otherwise it is 0. Indicates data d t Is it cached on edge server j? If data d is cached on edge server j, t ,but The value of is 1, otherwise it is 0, defining the unit cost of cache data on the edge server as cost; Step 3: Initialize the edge server coverage H, user set U, edge server set V, and select an edge server v from the edge server set V j , let the resource capacity of the server be c j ; Step 4: If the edge server v j User i in the coverage area has requested data d t demand, that is When the value of is 1, p t ←p t -s t cost, otherwise p t ←0, repeat step 4 until all data are traversed; Step 5: Use FPTAS to analyze the data d t The revenue p t After preprocessing, we get p′ t As the value of the data benefit actually used in the subsequent algorithm; Step 6: Use the dynamic programming algorithm to process the formalized data cache problem in step 2. Step 5 p′ t As data d t The benefit of caching to edge server j is to create a two-dimensional array B[t][m], which represents the highest benefit of loading the edge server with capacity m in the first t data; Step 7: Initialize B[t][m] and perform dynamic programming to solve it. The recursive formula of dynamic programming is defined as: Step 8: After the dynamic programming solution is completed, pruning operation is performed. If the data d t It has been cached on edge server j, that is, The value of is 1. At the same time, if there is user i within the coverage area of server j, then Ensure that in the next iteration, there is no need to process user i's data d t Requests; Step 9: Repeat steps 3 to 8 until the server has no remaining capacity or no other data needs to be cached, and then output the data cache solution S * And the corresponding data cache benefit R.
2. The data caching method based on popular content prediction in edge computing according to claim 1 is characterized in that: The data preprocessing is specifically to perform randomness and stationarity detection on the time series of historical data, so as to determine the parameters for constructing the ARIMA model.
3. The data caching method based on popular content prediction in edge computing according to claim 1 is characterized in that: The ARIMA model is written as ARIMA(p, d, q), which includes three parameters, where p is the autoregressive order, d is the number of differences, and q is the moving average order.
4. The data caching method based on popular content prediction in edge computing according to claim 1 is characterized in that: The randomness and stability tests are as follows: Step 11: Perform a stationarity test on the time series of historical data. If the p-value corresponding to the unit root test statistic is greater than 0.05, it is a non-stationary series and proceed to step 12; Step 12: Perform differential operation on the historical data time series until the time series becomes a stationary series; Step 13: Perform white noise test. If it is a white noise sequence, end directly. Otherwise, it means that the time series is a non-white noise sequence, and go to step 14; Step 14: Determine the model order of the processed stationary non-white noise sequence; Step 15: After the model is determined, the residual sequence needs to be tested for white noise. If the residual sequence is a white noise sequence, it means that there is no useful information in the residual, and the detection process ends. If it is a non-white noise sequence, repeat step 14.
5. The data caching method based on popular content prediction in edge computing according to claim 4 is characterized in that: Step 12 performs a differential operation and performs at most a second-order differential.
6. The data caching method based on popular content prediction in edge computing according to claim 4 is characterized in that: Step 13 performs a white noise test using the LB statistic, and then calculates the corresponding p-value. If the p-value is significantly greater than the significance level of 0.05, it is a white noise sequence.
7. The data caching method based on popular content prediction in edge computing according to claim 4 is characterized in that: In step 14, the maximum likelihood ratio method is used to estimate the parameters of the model and the values of each parameter are estimated. Then, for each different model, the BIC information criterion is used to determine the model order and determine the p and q parameters.
8. The data caching method based on popular content prediction in edge computing according to claim 1, characterized in that: In step 3, the formalized data caching problem constructed in step 2 is regarded as a dynamic multidimensional knapsack problem and solved using the MKP algorithm.
9. The data caching method based on popular content prediction in edge computing according to claim 1, characterized in that: In step 5, FPTAS is used to analyze the above data d t The revenue p t The specific preprocessing is as follows: Step 51: Take the maximum value of the above data benefits and set it as P, that is, P←max{p t |1≤t≤D}; Step 52: Let θ∈(0,1], let For all data d t ,make 10. The data caching method based on popular content prediction in edge computing according to claim 9, characterized in that: After step 5, the optimization problem has been transformed into a 0-1 knapsack problem, which is processed by the dynamic programming algorithm in step 6.
Citation Information
Patent Citations
Method for actively caching and replacing popular data in edge computing environment
CN112003921A
Social network short video placement method in mobile edge computing
CN114629928A
