A mobile edge network-oriented similar cache content dynamic placement method

Abstract

The application discloses a kind of similar cache content dynamic placement method for mobile edge network, similar cache is introduced in cooperative mobile edge cache, and similar content different from content request can be delivered.First, solve the cache placement problem of highly dynamic content popularity and network connection state in similar cache.Then a kind of similar cache placement algorithm and similar content delivery algorithm based on counter multi-arm tiger machine are disclosed;The problem is modeled as a counter multi-arm tiger machine model;Select a pull arm, call similar content delivery algorithm to process requests, collect the cache reward of the current stage, return to the similar cache placement algorithm, estimate the reward and weight of the arm according to the reward, repeat execution until termination, and the algorithm can obtain an excellent cache placement scheme adapted to the current environment.Through the mode, the application can solve the similar cooperative cache placement problem existing in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of network optimization technology, specifically relating to a method for dynamically placing similar cached content for mobile edge networks. Background Technology

[0002] In recent years, with the development of mobile internet technology, the increasing prevalence of 5G networks, and the proliferation of mobile devices, a series of emerging applications have arisen, such as the Internet of Things (IoT), the Internet of Vehicles (IoV), virtual reality, and augmented reality. These applications place higher demands on network throughput and stricter requirements on network latency, leading to an explosive growth in mobile traffic. To adapt to the ever-increasing mobile data traffic and alleviate the pressure on backhaul traffic, many researchers have focused on collaborative mobile edge caching systems that can fully utilize limited edge caching resources to provide services to users. The vast majority of these studies target traditional exact caching, where a cache hit occurs when the requested content is cached in the caching node; otherwise, a miss occurs, and the content is provided by a remote server. Because edge caching nodes have expensive and limited resources, and the edge caching environment is highly dynamic, exact caching frequently results in misses. User requests are then sent to the core network via the backhaul link, extending the user's service wait time and degrading the user's QoS.

[0003] In many cases, content similar to the user's requested content can be delivered to satisfy the user's request. For example, a user's request for high-resolution video can be satisfied by a low-resolution version. In other cases, the user's query is imprecise, meaning the request is for similar content. In these situations, the server can provide the user with one or more content approximations of the requested content, or forward the request to a remote server. This caching method, unlike traditional exact caching, is called similarity caching.

[0004] Many existing studies on similarity caching focus on single-caching scenarios, where a user's service is served by a single cache node, otherwise the request is forwarded to a remote server. Research on similarity caching in collaborative mobile edge caching scenarios is relatively limited. Considering the highly dynamic nature of content popularity in mobile edge caching, and the highly dynamic and variable network environment, an online learning algorithm is needed to address the similarity cache placement problem in collaborative edge caching, one that can learn from the environment and adapt to changes.

[0005] Adversarial multi-armed slots are a variant of the traditional slot machine problem. They differ from traditional random slots in several ways: random slots assume a constant reward distribution for each arm, aiming to find the arm with the highest average reward over time; in adversarial multi-armed slots, rewards are generated by opponents, and the probability of rewards can change at any time. The goal is to find the arm that produces high rewards even in the presence of adaptive opponents. Adversarial slots also feature arbitrary sequences of action rewards without statistical assumptions. This lack of statistical reward assumptions is more suitable for complex environments.

[0006] Existing caching strategies rely on prior assumptions or knowledge about user request sequences. However, in practical applications, the statistical characteristics of content popularity in mobile edge caching are often unknown, or even time-varying and non-stationary. Therefore, assuming known or predictable content request rates is impractical. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention provides a dynamic placement method for similar cached content in mobile edge networks, solving the cache placement problem faced by highly dynamic content popularity and network connectivity status in similar caches. First, a similar cache is introduced into the collaborative mobile edge cache, allowing delivery of similar content that differs from the requested content. Then, a similar cache placement algorithm and a similar content delivery algorithm based on an adversarial multi-armed slot machine model are disclosed. The problem is modeled as an adversarial multi-armed slot machine model; one arm is selected to be pulled, the similar content delivery algorithm is invoked to process the request, the cache reward for the current stage is collected, and the similar cache placement algorithm is returned. Based on the reward, the reward and weight of the arm are estimated, and this process is repeated until termination. The algorithm can obtain an excellent cache placement scheme adapted to the current environment. In this way, this invention can solve the problem of similar collaborative cache placement in existing technologies.

[0008] The technical solution adopted by the present invention to solve its technical problem includes the following steps:

[0009] Step 1: Build a similar caching system based on collaborative mobile edge caching;

[0010] Step 2: Construct a user request, latency cost model, and dissimilar cache cost model in the system to obtain the weighted cost of latency and dissimilar cost when serving user requests;

[0011] Step 3: Model the cache placement problem using an adversarial multi-armed slot machine, with the reward function set to the cost savings compared to fetching content from a cloud server, thus transforming the weighted cost problem of minimizing latency and dissimilarity costs into maximizing cumulative rewards;

[0012] Step 4: Based on the adversarial multi-armed slot machine similar cache placement algorithm, select one arm to pull, that is, determine the cache placement strategy, call the similar content delivery algorithm to process the request, collect the cache reward of the current stage, return to the similar cache placement algorithm, and estimate the reward and weight of each arm based on the reward.

[0013] Step 5: Repeat step 4 until termination, that is, serve all user requests set during the algorithm's runtime, and obtain an excellent cache placement scheme adapted to the current environment.

[0014] Further, step 1 specifically includes:

[0015] A set of cooperating base stations is included in the mobile edge cache. The base station connects to the core network via a backhaul link and accesses the cloud server to retrieve content from the database; all content is cached in the cloud server. This is equivalent to a content repository, meaning that user requests forwarded to the cloud server will be fulfilled, assuming that each piece of content f has the same size and is s. f ;

[0016] Assume the base stations have the same buffer capacity, C. S =k*s f That is, each base station can cache k pieces of content; the maximum distance for communication between base stations is D. BS If the distance d between base station i and base station j ij <D BS Then the two can communicate with each other and share cached content; using This represents a time series, where T is a finite time range, and caching decisions are updated periodically within each time period.

[0017] A central controller (CCU) is introduced in the collaborative area. It has the location and cache information of all terminals and BSSs, deploys cached content for base stations, and responds to user requests based on the current cache status to determine the content to be delivered.

[0018] Furthermore, step 2 specifically includes:

[0019] Step 2-1: Divide the request types into two categories K = {1, 2}: the first category is dissimilarity-sensitive requests, and the second category is latency-sensitive requests; in the two types of requests, dissimilarity cost and latency cost have different proportions, so introduce a weight w related to the dissimilarity cost. k , 0≤w k ≤1, the weight of delay cost is 1-w k ;

[0020] Based on the request type, an acceptable dissimilarity level β and an acceptable latency level δ are introduced for each request to further improve the user's QoS; if the dissimilarity cost of no content in the collaborative area is less than β and the transmission latency is less than δ, then the system will transmit the requested content to the user on the cloud server.

[0021] Step 2-2: The latency cost incurred in serving user requests at time t:

[0022]

[0023] in, For base station p k With p k+1 Transmission delay between;

[0024] Steps 2-3: Consider the case where the set F of content is a finite set. The dissimilarity cost is represented by a non-negative matrix |F|×|F|:

[0025]

[0026] Among them, C a (x,y) represents the cost of the difference between content x and content y. For all content x in F, C a (x,x)=0;

[0027] At time t, based on the user's request r, type k, acceptable dissimilarity β, and acceptable latency δ, the cost function for selecting r′ as the delivered content is as follows:

[0028] C (r,k,β,δ),r′ (X t ) = w k C a (r,r′)+(1-w k )D (r,k,β,δ),r′ (X t ).

[0029] Furthermore, step 3 specifically includes:

[0030] Step 3-1: The Central Control Unit (CCU) acts as a slot machine player, with different cache placement schemes for each arm. The reward function is set as reward = maxCost – cost, where maxCost represents the cost incurred when obtaining services from the cloud server. The total time is divided into T time periods. When selecting caching strategy X in time period t... t At that time, the service cost of each base station i is obtained. Then, the cost savings from the cache placement scheme were used as a reward metric for training against multi-armed slot machines;

[0031] Step 3-2: Users can obtain services from cached content stored in the local base station; in a collaborative base station environment, the requests of local users are met by the collaborative base station nodes or cloud servers based on the cost of obtaining the content.

[0032] Step 3-2-1: Assume the content requested by the user is available on the local base station i, and the content can be delivered immediately with low latency. The latency cost of delivering the content is d. i The number of user requests satisfied through local base station i at time t is The latency cost of local base station service is

[0033] Step 3-2-2: Assuming the content requested by the user is unavailable on the corresponding base station i, but available on other base stations in the cooperative area, the content is provided to the user via path p, and the transmission delay of path p is denoted as d. p The number of services obtained in the cooperative region at time t is denoted as . The latency cost of obtaining services within the collaborative region is

[0034] Step 3-2-3: Assuming no similar match occurs in the collaborative region, the user's request will be forwarded to the cloud server, and the corresponding content will be obtained from the cloud server. The transmission latency at this point is d. backhaul The number of services obtained in the cooperative region at time t is denoted as . Therefore, the cost of cloud server services is expressed as

[0035] Step 3-3: The total service latency cost in time t is:

[0036]

[0037] Considering similarity caching, similar content is delivered to users. If the delivered content differs from the user's requested content, a dissimilarity cost is incurred, which is included as a penalty cost in the reward. This cost only applies to content delivered within the collaborative region, and is expressed as:

[0038]

[0039] in, This indicates the number of similar hits that occurred within the collaborative region;

[0040] Steps 3-4: Indicates request r i The service type is weighted; an acceptable dissimilarity level β and an acceptable latency level δ are introduced for each request, and the cost function is as follows:

[0041]

[0042] The reward function is set to reward = maxCost – cost, as follows:

[0043]

[0044] Maximizing rewards means maximizing cost savings. The system's immediate reward is defined as follows:

[0045]

[0046] Further, step 4 specifically involves:

[0047] Based on the adversarial multi-armed slot machine similarity cache placement algorithm, considering the time flow as 1, 2, ..., t, ..., T, explore the set of arms. For the arm assembly Each arm x in the array maintains a weight and is initialized to w. x (1) = 1, in

[0048] At each time t, calculate the probability distribution for each arm. To explore factors, select a cache placement strategy, arm Then, the similar content delivery algorithm is invoked to process the request. Based on β and δ in the request, a candidate set of similar content in the collaborative region is determined. The cost of each piece of content in the set as delivery content is calculated, and the content with the lowest cost is selected as the delivery content. The delivery content is then delivered to the user, and the reward R for serving the user is calculated. t (X t );according to Estimated reward, based on Update the weights of the arm.

[0049] Furthermore, step 5 specifically includes:

[0050] Repeat step 4 until the end, and maintain the weight of the arm based on the observed reward. If the reward value of the arm is large, the weight of the arm will increase, and vice versa. By learning the environment, adapting to the dynamic changes of the environment, and adjusting over time, a superior cache placement scheme that is suitable for the current environment can be obtained.

[0051] The beneficial effects of this invention are as follows:

[0052] This invention solves the cache placement problem in collaborative mobile edge caching networks that allow the delivery of similar content. It can minimize the weighted cost of dissimilar costs incurred by service users in the system and transmission latency, that is, maximize the cumulative arm reward and find the best cache placement strategy that is suitable for the current environment. Attached Figure Description

[0053] Figure 1 This is a flowchart of the method of the present invention.

[0054] Figure 2 This is a model diagram of a similar caching system based on collaborative mobile edge caching in this invention.

[0055] Figure 3 This is a schematic diagram of the problem in this invention. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0057] This invention addresses existing problems by proposing a dynamic placement method for similar cached content in mobile edge networks. By modeling the similar cache placement problem in collaborative networks as an adversarial multi-armed slot machine model, it makes no assumptions about the stationarity of content popularity and considers changes in network state within the collaborative network. The model can learn and adapt to the environment through multiple iterations, adjusting to obtain an optimal cache placement scheme for the current environment. This solves the cache placement problem in collaborative mobile edge caching networks that allow the delivery of similar content. It minimizes the weighted cost of dissimilar costs incurred by users and transmission latency, i.e., maximizes the cumulative arm reward, and finds the optimal cache placement strategy for the current environment.

[0058] The objective of this invention can be achieved by adopting the following technical solutions:

[0059] like Figure 1 As shown, a method for dynamically placing similar cached content for mobile edge networks includes the following steps:

[0060] Step 1: As Figure 2 As shown, a similar caching system based on collaborative mobile edge caching is constructed;

[0061] Step 2: Construct a user request, latency cost model, and dissimilar cache cost model in the system to obtain the weighted cost of latency and dissimilar cost when serving user requests;

[0062] Step 3: As Figure 3As shown, the cache placement problem is modeled using an adversarial multi-armed slot machine, with the reward function set to the cost saved compared to fetching content from a cloud server, thus transforming the weighted cost problem of minimizing latency and dissimilarity costs into maximizing cumulative rewards.

[0063] Step 4: Based on the adversarial multi-armed slot machine similar cache placement algorithm, select one arm to pull, that is, determine the cache placement strategy, call the similar content delivery algorithm to process the request, collect the cache reward of the current stage, return to the similar cache placement algorithm, and estimate the reward and weight of each arm based on the reward.

[0064] Step 5: By repeating Step 4 until termination, the algorithm can adapt to the dynamic environment and adjust over time to obtain an excellent cache placement scheme that is suitable for the current environment. Specific implementation examples:

[0066] Step one specifically involves: constructing a similar caching system based on cooperative mobile edge caching. The mobile edge cache includes a set of cooperating base stations. The base station connects to the core network via a backhaul link and accesses the cloud server to retrieve content from the database. All content is cached on the cloud server. This is equivalent to a content repository, meaning that user requests forwarded to the cloud server will be fulfilled, assuming that each piece of content f has the same size and is s. f Assume that the base stations have the same buffer capacity, C. S =k*s f This means that each base station can cache k pieces of content. The maximum distance for communication between base stations is D. BS If the distance d between base station i and base station j ij <D BS Then the two can communicate with each other and share cached content. This represents a time series, where T is a finite time range, and caching decisions are updated periodically within each time period.

[0067] To facilitate problem research, a Central Control Unit (CCU) is introduced in the collaborative area. It has the location information and cache information of all terminals and BBSs, which can easily deploy cached content for base stations and respond to user requests based on the current cache status to determine the content to be delivered.

[0068] Step two involves constructing the entire caching system model.

[0069] First, request types are categorized into two classes K = {1, 2}: the first class is dissimilarity-sensitive requests, and the second class is latency-sensitive requests. Dissimilarity costs and latency costs have different proportions in the two types of requests, so a weight w related to the dissimilarity cost is introduced. k (0≤w k ≤1), the weight of delay cost is 1-w k An acceptable dissimilarity level β is introduced for each request based on its type. For dissimilarity-sensitive requests, the β is smaller than that for latency-sensitive requests. An acceptable latency level δ is introduced to further improve the user's QoS. If the dissimilarity cost of no content in the collaborative region is less than β and the transmission latency is less than δ, the system will transmit the requested content to the user via the cloud server. In summary, a request includes content r, request type k, tolerable dissimilarity level β, and acceptable latency level δ.

[0070] Secondly, at time t, the latency cost incurred in serving user requests:

[0071]

[0072] in, For base station p k With p k+1 The transmission delay between them.

[0073] Next, we consider the case where the set of content F is a finite set. Dissimilarity costs can be represented by a non-negative matrix of |F|×|F|.

[0074]

[0075] Among them, C a (x,y) represents the cost of the difference between content x and content y. For all content x in F, C a (x,x)=0.

[0076] The cost function for selecting r' as the content to be delivered at time t, based on the user's request r, type k, acceptable dissimilarity β, and acceptable latency v, is as follows:

[0077] C (r,k,β,δ),r′ (X t ) = w k C a (r,r′)+(1-w k )D (r,k,β,δ),r′ (X t )

[0078] Step three involves modeling the similar cache placement problem using an adversarial multi-armed slot machine. In this decision model, the CCU acts as the slot machine player, the arms represent different cache placement schemes, and the reward function is set as reward = maxCost – cost, where maxCost represents the cost incurred when obtaining services from the cloud server. The total time is divided into T time slots, each 1 ≤ t ≤ T being sufficient to complete the action and evaluation of the selected cache placement scheme. When cache strategy X is selected in time slot t... t At that time, we obtain the service cost for each base station i. The cost savings from using an efficient cache placement scheme are then used as a reward metric for training against multi-armed slot machines. Based on this information, the CCU decides which cache placement strategy to choose from a series of trials in order to maximize the cumulative reward.

[0079] Users can obtain services from cached content stored locally at the base station. In a collaborative base station environment, local user requests are fulfilled by collaborative base station nodes or cloud servers, depending on the content retrieval cost.

[0080] 1) Assume the content requested by the user is available on the local base station i. The content can be delivered immediately with low latency. The latency cost of delivering the content is d. i The number of user requests satisfied through local base station i at time t is . The latency cost of local base station service is

[0081] 2) Suppose that some content requested by a user is unavailable on the corresponding base station i, but is available on other base stations in the cooperative area. The content will then be provided to the user via path p. The transmission delay of path p is denoted as d. p The number of services obtained in the cooperative region at time t is denoted as . The latency cost of obtaining services within the collaborative region is

[0082] 3) Assuming no similar match occurs in the collaborative region, the user's request will be forwarded to the cloud server, and the corresponding content will be obtained from the cloud server. The transmission latency in this case is d. backhaul The number of services obtained in the cooperative region at time t is denoted as . Therefore, the cost of cloud server services is expressed as

[0083] The total service latency cost in time t is:

[0084]

[0085] Consider similarity caching, delivering similar content to users. If the delivered content differs from the user's requested content, a dissimilarity cost is incurred, which we include as a penalty cost in the reward. This cost only applies to content delivered within collaborative regions, and we represent it as follows:

[0086]

[0087] in, This indicates the number of times a similar hit occurs within the collaborative region.

[0088] The Content Delivery Control Unit (CCU) should design different content delivery strategies for different user request types to maximize user QoS. Since users have different interests, their content requests also fall into different types, and the CCU can deliver content that users are interested in. To achieve this, we categorize request types into two types K = {1, 2}: the first type is dissimilarity-sensitive requests, and the second type is latency-sensitive requests. Dissimilarity costs and latency costs have different proportions in the two types of requests. For the first type of request, we focus more on the impact of dissimilarity costs; therefore, we introduce a weight w related to dissimilarity costs. k In both types of requests, w k Different values ​​can be used (e.g., the request difference cost weight for type 1 is w1 = 0.8, and the request difference cost weight for type 2 is w2 = 0.3). Indicates request r i The service type is weighted; for each request, an acceptable dissimilarity level β and an acceptable latency level δ are introduced, and the cost function is as follows:

[0089]

[0090] An efficient caching strategy is needed to save costs. The reward function is set as reward = maxCost – cost, as follows:

[0091]

[0092] Maximizing rewards means maximizing cost savings. The system's immediate reward is defined as:

[0093]

[0094] In step four, the similar cache placement algorithm and similar content delivery algorithm based on adversarial multi-armed slot machine are used to place the cache strategy in the cooperative area and process the request to determine the content to be delivered.

[0095] Based on the adversarial multi-armed slot machine similarity cache placement algorithm, considering the time flow as 1, 2, ..., t, ..., T, explore the set of arms. For the arm assembly Each arm x in the array maintains a weight and is initialized to w. k (1) = 1, at each time t, calculate the probability distribution of each arm. To explore factors, select a cache placement strategy (arm) The similar content delivery algorithm is then invoked to process the request. Based on β and δ in the request, it determines a candidate set of similar content in the collaborative region. The cost of delivering each piece of content in the set is calculated, and the content with the lowest cost is selected as the delivered content. The delivered content is then delivered to the user, and the reward R for serving the user is calculated. t (X t ).according to Estimated reward, based on Update the weights of the arm.

[0096] Step five involves repeating step four until all user requests set during algorithm execution have been processed. The algorithm maintains the arm's weight based on observed rewards; if the arm receives a large reward, its weight increases, and vice versa. By learning from the environment and adapting to dynamic changes, the algorithm can adjust over time to obtain an optimal cache placement scheme suitable for the current environment.

Claims

1. A method for dynamically placing similar cached content in mobile edge networks, characterized in that, Includes the following steps: Step 1: Build a similar caching system based on collaborative mobile edge caching; Step 2: Construct a user request, latency cost model, and dissimilar cache cost model in the system to obtain the weighted cost of latency and dissimilar cost when serving user requests; Step 3: Model the cache placement problem using an adversarial multi-armed slot machine, with the reward function set to the cost savings compared to fetching content from a cloud server, thus transforming the weighted cost problem of minimizing latency and dissimilarity costs into maximizing cumulative rewards; Step 4: Based on the adversarial multi-armed slot machine similar cache placement algorithm, select one arm to pull, that is, determine the cache placement strategy, call the similar content delivery algorithm to process the request, collect the cache reward of the current stage, return to the similar cache placement algorithm, and estimate the reward and weight of each arm based on the reward. Step 5: Repeat step 4 until termination, that is, serve all user requests set during the algorithm's runtime, and obtain an excellent cache placement scheme adapted to the current environment.

2. The method for dynamically placing similar cached content for mobile edge networks according to claim 1, characterized in that, Step 1 specifically involves: A set of cooperating base stations is included in the mobile edge cache. The base station connects to the core network via a backhaul link and accesses the cloud server to retrieve content from the database; all content is cached in the cloud server. This is equivalent to a content repository, meaning that user requests forwarded to the cloud server will be fulfilled, assuming that each piece of content f has the same size and is s. f ; Assume the base stations have the same buffer capacity, C. S =k*s f That is, each base station can cache k pieces of content; the maximum distance for communication between base stations is D. BS If the distance d between base station i and base station j ij <D BS Then the two can communicate with each other and share cached content; using This represents a time series, where T is a finite time range, and caching decisions are updated periodically within each time period. A central controller (CCU) is introduced in the collaborative area. It has the location and cache information of all terminals and BSSs, deploys cached content for base stations, and responds to user requests based on the current cache status to determine the content to be delivered.

3. The method for dynamically placing similar cached content for mobile edge networks according to claim 2, characterized in that, Step 2 specifically involves: Step 2-1: Divide the request types into two categories K = {1, 2}: the first category is dissimilarity-sensitive requests, and the second category is latency-sensitive requests; in the two types of requests, dissimilarity cost and latency cost have different proportions, so introduce a weight w related to the dissimilarity cost. k , 0≤w k ≤1, the weight of delay cost is 1-w k ; Based on the request type, an acceptable dissimilarity level β and an acceptable latency level δ are introduced for each request to further improve the user's QoS; if the dissimilarity cost of no content in the collaborative area is less than β and the transmission latency is less than δ, then the system will transmit the requested content to the user on the cloud server. Step 2-2: The latency cost incurred in serving user requests at time t: in, d p <δ, For base station p k With p k+1 Transmission delay between; Steps 2-3: Consider the case where the set F of content is a finite set. The dissimilarity cost is represented by a non-negative matrix |F|×|F|: Among them, C a (x,y) represents the cost of the difference between content x and content y. For all content x in F, C a (x,x)=0; At time t, based on the user's request r, type k, acceptable dissimilarity β, and acceptable latency δ, the cost function for selecting r′ as the delivered content is as follows: C (r,k,β,δ),r′ (X t )=w k C a (r,r′)+(1-w k )D (r,k,β,δ),r′ (X t )。 4. A method for dynamically placing similar cached content for mobile edge networks according to claim 3, characterized in that, Step 3 specifically involves: Step 3-1: The Central Control Unit (CCU) acts as a slot machine player, with different cache placement schemes for each arm. The reward function is set as reward = maxCost – cost, where maxCost represents the cost incurred when obtaining services from the cloud server. The total time is divided into T time periods. When selecting caching strategy X in time period t... t At that time, the service cost of each base station i is obtained. Then, the cost savings from the cache placement scheme were used as a reward metric for training against multi-armed slot machines; Step 3-2: Users can obtain services from cached content stored in the local base station; in a collaborative base station environment, the requests of local users are met by the collaborative base station nodes or cloud servers based on the cost of obtaining the content. Step 3-2-1: Assume the content requested by the user is available on the local base station i, and the content can be delivered immediately with low latency. The latency cost of delivering the content is d. i The number of user requests satisfied through local base station i at time t is The latency cost of local base station service is Step 3-2-2: Assuming the content requested by the user is unavailable on the corresponding base station i, but available on other base stations in the cooperative area, the content is provided to the user via path p, and the transmission delay of path p is denoted as d. p The number of services obtained in the cooperative region at time t is denoted as . The latency cost of obtaining services within the collaborative region is Step 3-2-3: Assuming no similar match occurs in the collaborative region, the user's request will be forwarded to the cloud server, and the corresponding content will be obtained from the cloud server. The transmission latency at this point is d. backhaul The number of services obtained in the cooperative region at time t is denoted as . Therefore, the cost of cloud server services is expressed as Step 3-3: The total service latency cost in time t is: Considering similarity caching, similar content is delivered to users. If the delivered content differs from the user's requested content, a dissimilarity cost is incurred, which is included as a penalty cost in the reward. This cost only applies to content delivered within the collaborative region, and is expressed as: in, This indicates the number of similar hits that occurred within the collaborative region; Steps 3-4: Indicates request r i The service type is weighted; an acceptable dissimilarity level β and an acceptable latency level v are introduced for each request, and the cost function is as follows: The reward function is set to reward = maxCost – cost, as follows: Maximizing rewards means maximizing cost savings. The system's immediate reward is defined as follows:

5. A method for dynamically placing similar cached content for mobile edge networks according to claim 4, characterized in that, Step 4 specifically involves... Based on the adversarial multi-armed slot machine similarity cache placement algorithm, considering the time flow as 1, 2, ..., t, ..., T, explore the set of arms. For the arm assembly Each arm x in the array maintains a weight and is initialized to w. x (1) = 1, at each time t, calculate the probability distribution of each arm. γ∈[0,1] is the exploration factor; select a cache placement strategy; arm Then, the similar content delivery algorithm is invoked to process the request. Based on β and δ in the request, the candidate set of similar content in the collaborative area is determined. The cost of each content in the set as the delivery content is calculated, and the content with the lowest cost is selected as the delivery content. Deliver the content to the user and calculate the reward R for serving the user. t (X t );according to Estimated reward, based on Update the weights of the arm.

6. A method for dynamically placing similar cached content for mobile edge networks according to claim 5, characterized in that, Step 5 specifically involves: Repeat step 4 until the end, and maintain the weight of the arm based on the observed reward. If the reward value of the arm is large, the weight of the arm will increase, and vice versa. By learning the environment, adapting to the dynamic changes of the environment, and adjusting over time, a superior cache placement scheme that is suitable for the current environment can be obtained.

Images