A method and system for collaborative optimization of a multi-tier edge cache network

Abstract

The application discloses a kind of multilayer edge cache network cooperative optimization method and system, it is related to edge computing and wireless communication technical field, this method considers the freshness (information age) of content, cache update cost and the dynamic change of user request simultaneously in content cache and update process, propose a new cache timeliness reward model, build a collaborative caching mechanism integrated with global optimization and local adaptive adjustment.The optimal update frequency of content is determined based on long-term popularity prediction on the MBS side to reduce update redundancy and ensure the timeliness of content;SBS side uses a dynamic adjustment strategy combining Lyapunov optimization and popularity prediction to maintain system stability while optimizing cache content in real time, improving the accuracy and timeliness of user content access.This method effectively improves the service quality of content distribution in edge network, reduces the cache update cost, and is suitable for vehicle networking, smart city and other scenarios with high information timeliness requirements.

Description

Technical Field

[0001] This invention relates to the fields of edge computing and wireless communication technology, and in particular to a method and system for collaborative optimization of multi-layer edge caching networks. Background Technology

[0002] With the rapid development of mobile internet and IoT technologies, edge caching networks have attracted widespread attention as an important means to improve content distribution efficiency and reduce access latency. In typical edge caching scenarios, infrastructure nodes (such as MBS, SBS, and drone nodes) cache content files that users may request, assisting mobile terminal devices in obtaining the required data in real time and efficiently, thereby significantly improving the system's service quality and user experience.

[0003] However, content data in edge caching networks typically exhibits significant time-sensitivity, with some environmental information or dynamic data losing its original application value within a short period. Traditional caching strategies largely rely on historical content popularity, neglecting the decay of data value over time. This strategy easily leads to outdated cached content, reducing the accuracy and timeliness of information retrieval for users and severely impacting overall system performance. Therefore, effectively incorporating content freshness considerations into caching strategy design has become an important research direction.

[0004] To address this, the academic community proposed the concept of Age of Information (AoI) to measure the freshness of cached content. AoI-based caching optimization methods aim to maximize content freshness under limited resource conditions, while balancing system transmission overhead and communication costs. However, in practical edge networks, content collection, caching, and updating consume significant wireless communication resources. While excessively high update frequencies can ensure content freshness, they significantly increase system load and energy consumption; conversely, excessively low update frequencies lead to outdated information, failing to meet users' demands for real-time performance and accuracy. Therefore, caching mechanism design must achieve an effective trade-off between content freshness and update costs.

[0005] Furthermore, user request behavior in edge caching networks is highly dynamic, and the distribution of content popularity changes constantly over time, making it difficult for a single static caching strategy to adapt to these changes. How to combine content popularity prediction with dynamic caching optimization to adjust caching strategies in real time to adapt to changes in user needs and further improve cache hit rate and content freshness is another challenge that urgently needs to be addressed in caching optimization research.

[0006] To address the aforementioned issues, current technologies lack an efficient caching optimization method that can simultaneously consider content freshness, update costs, and dynamic changes in popularity. This is particularly true in multi-layered network architectures (such as MBS and SBS collaboration), where a systematic mechanism for coordinating caching strategies across different layers to improve overall system performance is lacking. Therefore, there is an urgent need to propose a multi-layered network collaborative caching optimization method that combines content timeliness modeling, dynamic popularity prediction, and update cost control to achieve freshness-aware, efficient, and sustainable content management and distribution in edge caching networks. Summary of the Invention

[0007] This invention aims to address the performance degradation problem in existing edge caching networks caused by neglecting the timeliness of content data, while simultaneously balancing the conflict between cache update frequency and communication resource consumption. To address the issue that traditional caching mechanisms do not adequately consider information freshness and dynamic changes in user requests, this invention proposes a multi-layer edge caching network collaborative optimization method and system, which can significantly improve the freshness of content services and user experience with limited update overhead. Specifically, this invention achieves the above objectives through the following technical solutions:

[0008] Firstly, this application provides a multi-layer edge caching network collaborative optimization system, comprising multiple content providing devices (CDs), macro base stations (MBS), and multiple small base stations (SBS). In this network, the CDs are responsible for generating and uploading content to the MBS, while the MBS is responsible for distributing the content to the SBS for caching and storage. The SBS provides real-time content to the user end through local cache update decisions. At the user end, when a user needs to access certain content, the system provides the service from the most suitable node based on the user request, network conditions, and the timeliness of the cached content, ensuring the timeliness and reliability of the content. To comprehensively consider the transmission costs of each link in the network, this invention mainly considers the following two aspects of transmission costs: the transmission cost of CDs uploading content to the MBS, and the transmission cost of MBS providing content updates to the SBS. Considering these transmission costs helps to optimize the resource allocation and scheduling strategy of the overall caching network.

[0009] Based on the aforementioned network architecture, this invention introduces Information Age (AoI) as a key indicator to quantify and optimize the timeliness of cached content. AoI is an important metric for measuring the timeliness of cached content updates, representing the "obsolescence" of content from its generation or update at the source node to the current moment. A higher information age indicates lower timeliness and diminishing value of the content. In multi-layered, heterogeneous networks, information age can effectively measure the "obsolescence" of cached content, providing a theoretical basis for optimizing caching strategies.

[0010] This invention designs a novel content caching timeliness reward model. This model, combining the characteristics of information age, quantifies the value decay of cached content over time and guides the dynamic adjustment of cache update decisions based on this. Specifically, as time passes, the timeliness of content gradually decreases. The system needs to adjust its caching strategy promptly based on the freshness of the content to avoid outdated content occupying valuable cache resources, thereby improving the utilization efficiency of cache space. By introducing a dynamic adjustment mechanism for information age and update strategy, this invention can achieve precise optimization of cache timeliness, significantly improving the content transmission quality and service response speed in edge computing networks.

[0011] Ultimately, MBS and SBS are used to collaboratively execute a multi-layer edge caching network collaborative optimization method to achieve collaborative scheduling of global and local cache update decision optimization in multi-layer edge caching networks.

[0012] Secondly, this application also provides a multi-layer edge caching network collaborative optimization method, which includes an MBS caching optimization stage and an SBS caching optimization stage. The implementation process of each stage is described below.

[0013] 1) MBS cache optimization phase

[0014] MBS is the core node in the edge caching network, responsible for collecting and caching content files from CDs, and then providing them to SBSs within the region. The goal of MBS's caching optimization algorithm is to determine the frequency at which content files are updated from CDs in order to optimize the timeliness of content caching.

[0015] The optimization problem in the first phase can be described by maximizing the timeliness reward of MBS cached content files. Specifically, the objective is to maximize the timeliness reward of content, which is calculated using a popularity-weighted approach while imposing constraints on transmission costs. In this process, for each content file's popularity and the update strategy within each time slot, the timeliness reward is calculated by measuring the effectiveness of the update strategy through average timeliness over a finite time period.

[0016] To solve this optimization problem, time is first divided into multiple update intervals, with the timeliness reward for each interval determined by its length. By optimizing the number of updates for each content file, the problem is further decomposed into independent subproblems, and the optimal update interval is solved for each content file separately.

[0017] During the solution process, the optimal solution for each update interval can be obtained using the KKT (Karush-Kuhn-Tucker) conditions. After determining the number of updates, the optimal update interval for each content file will be equal, meaning that the update time interval for all content files is the same. By solving for the update interval, the optimization problem is transformed into solving for the update frequency of each content file.

[0018] In this optimization phase, the objective is to maximize the overall timeliness reward of MBS cached content. To achieve this, a Sequential Least Squares Programming (SLSQP) algorithm is used to solve for the optimal update frequency for each content file. Specifically, SLSQP is a constrained optimization algorithm that optimizes the update frequency under constraints by minimizing or maximizing the objective function. In this invention, the SLSQP method is used to solve how to maximize the timeliness reward of MBS cached content given transmission resources and update frequency constraints.

[0019] 2) SBS caching optimization phase

[0020] The second phase focuses on optimizing local cache update decisions within the region using the SBS. Unlike the global optimization of the MBS in the first phase, the SBS needs to dynamically respond to user requests within a shorter timescale, while balancing cache timeliness with update costs. Therefore, the core objective of the second phase is to dynamically adjust cached content to maximize the timeliness reward for users accessing content, while meeting update cost constraints.

[0021] To address the SBS cache update problem, two main factors are considered: the timeliness reward for users accessing content and the transmission cost for SBS to retrieve content from MBS. Based on these factors, a dynamic optimization framework for SBS caching is constructed using the Lyapunov optimization method. This framework introduces a virtual queue to constrain the frequency of cache updates, thereby balancing update costs and timeliness rewards.

[0022] First, the optimization objective is to maximize the timeliness reward of SBS cached content and optimize the system's cache update decisions over the long term. To control the update frequency and ensure manageable resource overhead, constraints are set to limit the number of updates within each time slot, and a virtual queue is used to control the content update frequency, thereby ensuring system stability.

[0023] In the Lyapunov optimization framework, a virtual queue is introduced to measure the pressure of cache updates. An increased virtual queue indicates that cache updates are too frequent, potentially exceeding the system's cost budget, while a stable virtual queue indicates that the update process is effectively controlled. In this way, the system can adaptively adjust its cache update strategy, optimizing decisions based on the current state in each time slot.

[0024] To make the optimal update decision within each time slot, the Drift-Plus-Penalty (DPP) method is employed. The goal of this method is to minimize the drift and penalty terms in each time slot, thereby maximizing system stability and timeliness rewards. During the solution process, the Lyapunov virtual queue is dynamically adjusted, and the update decision for the current time slot is obtained by minimizing the upper bound of the DPP.

[0025] Because SBS cache updates involve dynamic changes in user requests, simply predicting request counts based on historical data may not accurately capture request fluctuations. Therefore, the prediction target is shifted from specific request counts to predicting popularity distribution. Popularity distribution comprehensively reflects the popularity of content files and helps improve prediction accuracy. Based on this prediction, the timeliness bonus of content files can be calculated for each time slot, guiding cache update decisions.

[0026] Based on the predicted popularity distribution and timeliness rewards, SBS uses the Lyapunov optimization framework to solve the cache update decision for each time slot. This process adjusts the frequency of content updates by minimizing the upper bound of the Distributed Popularity Per Window (DPP) and ensures the system remains stable in the long term. Ultimately, SBS executes content update operations based on the optimized cache update decisions, thereby improving the freshness of cached content and the overall service quality of the system.

[0027] Throughout the optimization process, there is a trade-off between the stability and timeliness rewards of the Lyapunov virtual queue. By adjusting the control parameter V in the Lyapunov optimization, the system's performance and stability can be flexibly balanced. When the control parameter is large, the system tends to optimize timeliness rewards; when the control parameter is small, the system focuses more on controlling update costs, thereby achieving effective control of resource consumption.

[0028] 3) Two-stage collaborative caching optimization mechanism

[0029] To improve the overall performance of multi-layer edge caching networks and ensure that users can obtain timely and high-time-sensitivity content files, this invention proposes a layered, collaborative, two-stage caching optimization mechanism. This mechanism combines global and local optimization, utilizing different time scales for collaborative optimization to achieve the optimal cache update strategy.

[0030] Specifically, the first phase involves MBS performing global cache optimization. The optimization goal is to control the update frequency and interval of content files over a longer timescale, thereby ensuring the long-term effectiveness of the globally cached content. The second phase involves SBS performing local cache optimization. Over a shorter timescale, SBS dynamically adjusts its caching strategy based on the cached content information and local request data provided by MBS in the first phase, optimizing the quality of content service for users. During this collaborative optimization process, MBS and SBS are linked through popularity prediction to effectively integrate long-term global optimization with real-time local optimization.

[0031] In the first phase, MBS's task is to optimize cache update decisions based on a longer time scale (interval) to ensure that cached content remains efficient and timely over the long term. At the beginning of each interval, MBS first analyzes the content request data of all SBSs in the previous time period and extracts the global popularity distribution (historical popularity distribution). Based on this historical data, MBS uses a trained LSTM neural network model to predict the global popularity distribution for the current time period. Once the popularity distribution is predicted, MBS combines this information with update frequency constraints and employs the SLSQP optimization algorithm to calculate the optimal update frequency and update interval for each content file. This optimization process ensures that MBS can balance update frequency and update interval globally to maximize long-term timeliness rewards, thereby optimizing the overall caching performance of the network.

[0032] In the second phase, SBS adjusts its local cache update decisions based on global cache information provided by MBS and its own real-time local request data. Within each shorter timescale, SBS first analyzes the content request data from the previous period to obtain the local popularity distribution. Based on this data, SBS uses an LSTM neural network model to predict the current period's request popularity trend. Then, SBS combines the popularity prediction results with the Lyapunov optimization framework, using virtual queues to control the frequency of content updates, ensuring the system maximizes timeliness rewards while controlling update costs. Specifically, SBS uses the DPP method to minimize system drift and penalty terms, thereby dynamically determining the content files to be updated in each time slot based on real-time feedback.

[0033] In this optimization process, SBS's decisions rely on real-time request data and MBS's global cache information. Through this dynamic adjustment mechanism, SBS can achieve immediate response to user needs while ensuring controllable update costs, thereby improving the quality of content services for users. The two-stage collaborative caching optimization mechanism can effectively improve the overall performance of the multi-layer edge caching network and ensure optimal cache update decisions at different time scales, meeting users' needs for highly time-sensitive content.

[0034] Compared with the prior art, the beneficial technical effects of the present invention are:

[0035] 1. Establish a content timeliness reward model based on information age to compensate for the problem that traditional caching mechanisms ignore content timeliness;

[0036] 2. Based on popularity prediction and dynamic optimization, proactive updates and intelligent adjustments to cached content are achieved;

[0037] 3. A two-stage collaborative optimization mechanism is used to balance long-term global performance with local real-time response;

[0038] 4. Under the constraints of limited update frequency and communication resources, significantly improve the timeliness of content access for users and the overall service quality of the system;

[0039] 5. It improves the operating efficiency and resource utilization of edge caching networks, has good application value, and is suitable for scenarios with high requirements for information timeliness, such as vehicle networking and smart cities. Attached Figure Description

[0040] Figure 1 This is a scenario diagram of the multi-layer edge caching network collaborative optimization system provided in this application;

[0041] Figure 2 This is a flowchart of the multi-layer edge caching network collaborative optimization method provided in this application, wherein: Figure 2 .1 is a flowchart of the two-stage collaborative caching optimization mechanism. Figure 2 .2 is the MBS cache optimization phase. Figure 2 .3 is the SBS cache optimization stage;

[0042] Figure 3 This is a timeliness caching curve of the content files provided in this application;

[0043] Figure 4 The image provided in Simulation Example 1 shows the MBS cache optimization stage, where: (a) is the average AoI histogram of all cached content files on the MBS side, and (b) is the average timeliness histogram of all cached content files on the MBS side.

[0044] Figure 5The images provided in Simulation Example 2 are images of the SBS cache optimization stage, where: (a) is a comparison chart of the global average timeliness of all cached content files on the SBS side, and (b) is a comparison chart of the average user request reward on the SBS side. Detailed Implementation

[0045] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0046] One embodiment of this application provides a multi-layer edge caching network collaborative optimization system. This system includes multiple content providing devices (CDs), macro base stations (MBSs), and multiple small base stations (SBSs) to jointly construct a multi-layer edge caching collaborative architecture, such as... Figure 1 As shown, the MBS node is deployed at the center of the network and is mainly responsible for global content popularity modeling and cache update decision-making. It distributes content files generated by CDs within the region to SBS nodes within the region for caching and storage based on cache update decisions. Multiple SBS nodes are evenly deployed within the MBS coverage area, periodically receiving content updates from the MBS, dynamically optimizing cache update decisions based on local real-time user requests, and providing real-time caching services to surrounding users. CD nodes are evenly distributed in the network environment, periodically uploading content update data to the MBS nodes to support the overall content update mechanism. MBS and SBS are used to collaboratively execute a multi-layer edge caching network collaborative optimization method to achieve collaborative scheduling of global and local cache update decision optimization in the multi-layer edge caching network.

[0047] In this embodiment, the distribution of user requests for content files follows a Zipf distribution, with the initial distribution parameter set to 1 to reflect user content access preferences. Within each short period, the number of user requests follows a Poisson distribution with a mean of 20, simulating the characteristics of random request arrival.

[0048] Based on the same inventive concept, another embodiment of this application provides a multi-layer edge caching network collaborative optimization method, which employs the following... Figure 2 The two-stage collaborative caching optimization mechanism shown in Figure 1 includes an MBS caching optimization stage and an SBS caching optimization stage. When a client requests content from the SBS, the two stages perform the following operations respectively:

[0049] Combination Figure 2 As shown in Figure 2, on the MBS side, based on the content popularity distribution prediction results, the optimal update frequency and interval for each content file are determined to maximize the timeliness reward of long-term cached content files under the constraint of transmission cost. Combined with... Figure 2As shown in Figure 3, on the SBS side, the timeliness reward for short-term cached content files is calculated based on the global cache update decision and content popularity distribution prediction results obtained from the MBS side. Under the Lyapunov optimization framework, a virtual queue is introduced, and the objective function of the drift plus penalty DPP term is used to dynamically control the local cache update decision for content files in each time slot, so as to balance the cache update cost and the timeliness reward for short-term cached content files. Based on the global cache update decision and the local cache update decision, the cached content on the MBS side and the SBS side is updated respectively in each time period, improving the timeliness of cached content in the overall network and responding to user content requests.

[0050] In this embodiment, the timeliness reward on the MBS side and SBS side is determined based on the content's information age (AoI) and the user's perception function of the content's timeliness. Specifically, for each content file, an AoI value is defined on the Content Provider Device (CD), MBS, and SBS, denoted as A... c,f (t), A m,f (t) and A u,f (t), where the AoI at the content generation end is initialized to 1. When MBS or SBS updates the content file from the parent node, its corresponding AoI value is reset to the AoI at the source end; otherwise, it increases linearly over time. A maximum information age is introduced. This serves as a content expiration threshold to determine whether content is still valid. Optionally, the system pre-sets a set of content files, with a total number F of 50 files, each of the same size. To reflect the differences in content importance and timeliness, all content files are divided into three categories according to predefined standards, each corresponding to a different maximum allowable information age. The first 20 files are set Set to 20, with the middle 15 files configured. Set to 40, with 15 files remaining. It is 80.

[0051] To characterize users' subjective perception of content timeliness, a perception function is introduced as the timeliness reward function T(A). This function is a variant of the Sigmoid function. The timeliness reward function T(A) constructed based on the AoI value and the perception function is defined as follows under different content expiration thresholds:

[0052]

[0053] Among them, when the content AoI is low, the timeliness reward decreases slowly, and users have a strong perception of the freshness of the content; when the AoI is in the middle range, the reward decreases rapidly; and when the AoI is close to the maximum threshold, the reward function slowly approaches zero, reflecting the process of users' perception of the value of outdated content rapidly decreasing and then stabilizing.

[0054] In real-world scenarios, different types of content files may have different validity periods. Therefore, the system further considers three different maximum AoI values, namely: This led to the establishment of a corresponding timeliness model, such as Figure 3 As shown.

[0055] In both phases, to achieve early detection of content request trends, the system predicts the popularity of content files based on historical request data and optimizes the caching update strategy of MBS / SBS ​​nodes accordingly. Specifically, the content popularity distribution prediction is implemented using an LSTM neural network model to capture the temporal correlation of content request sequences. The LSTM model consists of two LSTM layers and a fully connected output layer, and uses the Adam optimizer for parameter updates, with mean squared error (MSE) as the loss function during training.

[0056] The prediction process on the MBS side includes: within each long-term interval, the MBS node summarizes the content request data reported by each SBS node in the previous period and statistically analyzes the historical content request distribution. The data is then fed into a trained LSTM model, which outputs the predicted content popularity distribution for the current period. Used to guide MBS content collection optimization decisions. θ m,f This represents the relative request frequency of content file f in the previous period. The prediction process on this side is expressed as:

[0057]

[0058] Among them, f w (·) represents the trained LSTM prediction model, and w represents the neural network weight parameters. The training samples of the LSTM model come from the request log data accumulated during system operation. During training, MSE is used as the loss function to minimize the deviation between the predicted popularity and the actual popularity.

[0059] After obtaining the content popularity distribution prediction results for the current period Subsequently, MBS optimizes the update frequency and interval of each content file based on the prediction results to maximize the timeliness reward of long-term cached content files under the constraint of transmission cost. Specifically, this includes:

[0060] The MBS content caching algorithm determines the frequency of updating content files from CDs, optimizing the timeliness of content files retrieved by MBS. MBS (t). The optimization problem in the first stage can then be described as:

[0061]

[0062] The optimization problem is to determine the optimal timeliness of cached content files under the constraint of transmission cost. For T MBS (t), using the timeliness reward model T(A), represents the popularity-weighted timeliness reward. For the transmission cost C of MBS retrieving content files from CDs... MBS (t), to simplify the problem description, a fixed update frequency is used to represent the constraint of transmission cost. Therefore, the optimization problem in the first stage is transformed into maximizing the timeliness reward of content files in MBS under a fixed number of updates.

[0063] use Let f be an indicator that the MBS updates content file f from the CD within time slot t. Therefore, the update decision variable for MBS updating content file f can be expressed as: Then based on the given u m,f The average timeliness reward within time period T can be calculated as follows:

[0064]

[0065] Among them, T(A) m,f (t) represents the timeliness bonus for each content file f on the MBS side.

[0066] Content popularity distribution prediction results based on MBS current period interval The timeliness bonus for MBS cached content files is calculated as follows:

[0067]

[0068] For ease of calculation, time T is expressed as the update time interval τ. f,i The cache is divided into sections, and the timeliness bonus of the cache is evaluated within each update interval. For content file f, assume that N can be updated within time T. f Next, the optimal update strategy is solved. Given that the MBS update content file f has a time slot of... The update interval of content file f can be obtained:

[0069]

[0070] Further calculate the timeliness reward within each update interval, sum the timeliness rewards across all intervals within time T, and calculate the average timeliness reward for content files under different content expiration thresholds:

[0071]

[0072] Combining the predicted MBS content popularity distribution The MBS cache optimization problem is modeled as a function of the update interval τ.f The optimization problem is represented as:

[0073]

[0074] Solving the optimization problem in the first stage requires determining two key decision variables: the number of times N is updated for each content file f. f and the corresponding update time interval τ f Once N is determined f Then, combined with the predicted MBS content popularity distribution The MBS caching optimization problem can be further decomposed into independent subproblems for each content file, so that the optimal update interval for each content file can be solved separately. Therefore, given the number of updates N for content file f... f Under this premise, its sub-problems are modeled as follows:

[0075]

[0076] In the solution process, the optimal update interval for each content file is obtained using the KKT conditions, and the optimal update frequency for each content file is solved using a sequential quadratic programming algorithm. Specifically, since the subproblems of the content files are convex optimization problems, the KKT conditions can be used to solve them, and the optimal update interval for content file f is obtained as follows: and update frequency is This demonstrates that, given a limited number of updates, maintaining the same update interval is the optimal strategy. Substituting the update frequency into the equation yields different content expiration thresholds. The average timeliness bonus is:

[0077]

[0078] After obtaining the optimal update interval, the MBS cache optimization problem is transformed into solving for the update frequency of each content file, that is, using a sequential quadratic programming algorithm to solve for the optimal update frequency λ of each content file. f The cache update decision for content files is defined as the update frequency vector λ = [λ1, ..., λ]. F The optimization objective is to maximize the average timeliness reward of cached content files, expressed as:

[0079]

[0080] Where, λ m The total update frequency constraint for MBS to obtain content files from the previous node device within each time slot is the proportion of cached files updated in each time slot to the total number of content files.

[0081] Through the above optimizations, the system can dynamically adjust the update strategy of MBS cached content while reasonably controlling the content collection cost, providing fresh data support for subsequent local cache update decisions.

[0082] Following the MBS cache optimization in the first phase, the second phase involves further adjusting the SBS cache content update strategy based on the MBS optimization results. Compared to the global optimization strategy of MBS, SBS needs to dynamically respond to user requests on a shorter time scale, while balancing the timeliness of user request caching with the cost of content file updates. Therefore, the core objective of the second phase is to dynamically adjust cached content while meeting update cost constraints, maximizing the timeliness reward for users to obtain content files.

[0083] For SBS cache updates, two factors need to be considered: 1) the cache timeliness reward T obtained by the user. SBS (t); 2) Transmission cost C for SBS to obtain content files from MBS SBS (t). The optimization problem in the second stage can then be described as:

[0084]

[0085] Specifically, set This represents the number of requests for content file f received by the u-th node on the SBS side within time slot t. The AoI and timeliness reward of the content file at the SBS are defined as T(AoI, A ... u,f (t)). Therefore, the timeliness bonus of the content file obtained by the SBS side within time slot t can be expressed as:

[0086]

[0087] To make more accurate update decisions, SBS needs to predict the request volume of content files. To achieve this, an LSTM-based neural network model is also used. The content popularity distribution prediction process on the SBS side includes: at the beginning of each cache update cycle, statistically analyzing the content request data from the previous cycle to form a local popularity distribution. The data is then fed into a trained LSTM model, which outputs the predicted content popularity distribution for the current period. This prediction helps SBS accurately select content to update, guiding current cache update decisions.

[0088] To measure the timeliness benefit of updating content file f in the current time slot t, this benefit is defined as the difference in timeliness reward for the content file on the MBS side and the SBS side. This is based on the content popularity distribution prediction results in the current time slot t of the SBS. The weighted timeliness bonus of content file f within time slot t is further calculated as follows:

[0089]

[0090] in This is the predicted popularity value of SBS-side content file f. The final result will be obtained from this formula. Substitute the following process to further optimize the decision variables for updating content files on the SBS side.

[0091] Based on the above definition, a dynamic optimization problem for SBS caching is constructed using the Lyapunov optimization method. By introducing a virtual queue to constrain the cache update frequency, the following optimization framework is built to solve for the decision variables of SBS-side content file updates.

[0092]

[0093] in, v is the weighted timeliness reward for the content file f obtained by the u-th node on the SBS side within time slot t; v is the upper limit of the average update cost per time slot. It involves updating decision variables.

[0094] This section introduces the following Lyapunov virtual queue Q[t] to measure the cost pressure of SBS cache updates. Specifically, the virtual queue is updated in slot t as follows:

[0095] Q[t+1] = max{Q[t] + a[t] - v, 0}

[0096] in, This indicates the total number of content files updated by SBS in time slot t. The growth of this virtual queue indicates that cache updates are too frequent and may exceed the system's update cost budget, while queue convergence indicates that update costs are effectively controlled.

[0097] To dynamically adjust the cache update strategy, the drift-penalty DPP term from Lyapunov optimization is adopted, and its objective function is expressed as:

[0098] Δ(Q[t])+V·p(t)

[0099] Where Δ(Q[t])=L(Q[t+1])-L(Q[t]) represents the Lyapunov drift, Here, p(t) is the Lyapunov function used to measure the severity of queue backlog, p(t) is the system's immediate penalty function, corresponding to negative time-based rewards (i.e., negative utility), and V is an adjustable control parameter used to balance the relationship between time-based rewards and queue stability. A larger V tends to increase time-based rewards, while a smaller V focuses more on maintaining the stability of queue update costs.

[0100] With the goal of maximizing system stability and timeliness rewards, the DPP objective function is minimized at each time slot t to obtain the SBS content file update decision variables. Specifically, the upper bound of the DPP term's drift is established as follows: Let constant Then DPP can be simplified to:

[0101]

[0102] in This represents the expected time-sensitive reward. Ultimately, SBS updates the decision variables by solving the following minimization problem at each time slot t:

[0103]

[0104] The above method first initializes the cached content files in the SBS. Each time slot randomly retrieves *s* content files from the MBS at a fixed update frequency, until all content files have been cached once. In the initial stage of each time slot, the popularity distribution is calculated based on the requests from the previous time slot, and this distribution is used to predict the popularity distribution of the current time slot. Time-sensitive rewards obtained from updating content files Minimize the upper bound of the DPP to obtain the updated decision variables. After completing the update decision, the system... Determine the number of content files to be updated, a[t], and update the Lyapunov virtual queue Q[t+1]. The updated virtual queue is used to measure the system load and ensure that the system does not exceed the cost budget.

[0105] The two stages mentioned above coordinate based on the content popularity prediction results. MBS is responsible for global long-term cache optimization, while SBS is responsible for local real-time cache optimization, thereby achieving coordination between global and local, long-term and short-term cache decisions and improving the overall performance of the multi-layer edge cache network.

[0106] To verify the effectiveness of the content collection optimization mechanism in the MBS caching optimization stage proposed in this invention, this embodiment evaluates the MBS caching performance through simulation experiments, mainly analyzing it from two dimensions: average AoI and timeliness of content files. In the experiment, the MBS cached content is periodically replaced according to the optimal update frequency determined by the optimization algorithm. In the experimental settings, the total content update rate λ of MBS is... m The value is set to 10, meaning that a maximum of 10 content files are updated from the CD within each time period. This is used to simulate the cache optimization process under a limited update budget and to verify the effectiveness and adaptability of the algorithm. Figure 4 The average AoI and timeliness of all cached content files are displayed over 20 time periods. Figure 4(a) shows that the AoI of most content files remains around 2, with the overall range between 1 and 10. Figure 4 (b) Further demonstrates the average timeliness of all content files, which has remained stable at around 1 for a long time, indicating that cached content still maintains good real-time value and service performance.

[0107] In summary, the simulation results fully verify that the MBS content update mechanism proposed in this invention can significantly reduce the AoI of cached content and maintain high content timeliness under limited update resources, effectively improving the ability of downstream edge nodes to obtain high-quality content and providing stable support for the entire collaborative caching system.

[0108] To further verify the effectiveness of the SBS caching optimization mechanism proposed in this invention, this embodiment designs a set of simulation experiments to compare and analyze the performance of different algorithms in the content cache update process. The evaluation indicators include content-weighted timeliness and user request reward, where content-weighted timeliness refers to the globally weighted timeliness obtained after weighting the timeliness reward of the content file by predicted popularity. This reflects the freshness of content and user experience throughout the caching system. User request rewards. This means calculating the actual timeliness of content file retrieval by the user, which more accurately indicates the timeliness of the content received by the user. The comparison algorithms used are traditional algorithms: IUB (Informed Upper Bound), LFU (Least Frequently Used), and Myopic. The IUB algorithm assumes that the system can fully predict future request distribution, serving as a reference for theoretically optimal performance; the LFU algorithm selects cached update content based on historical request frequency; and the Myopic algorithm optimizes caching decisions only based on request information in the current time slot, ignoring long-term performance.

[0109] In the experimental setup, the average update cost parameter v = 2 was set during the SBS cache optimization process, and the control parameter V = 1000 in the Lyapunov optimization algorithm. Figure 5 (a) shows the changing trend of the globally weighted content timeliness over multiple time periods. The SBS caching optimization mechanism proposed in this invention quickly stabilizes above 0.80 after two time periods, which is better than other comparison algorithms and demonstrates a strong ability to maintain content freshness. Figure 5(b) illustrates the variation in the average request reward, an metric that measures the average timeliness of the content file corresponding to each serviced request, directly reflecting the user's perceived service quality. As shown in the figure, the algorithm of this invention maintains a consistently high level, with an average request reward of approximately 0.8, outperforming other comparative algorithms. The trends of weighted content timeliness and average request reward are similar in the figure, indicating that the LSTM neural network used to predict the popularity of content files closely resembles real user requests and can effectively reflect user requests.

[0110] In summary, the SBS caching mechanism proposed in this invention effectively maintains the high timeliness of content caching while ensuring the quality of user request response. By allowing for a moderate increase in update costs and dynamically balancing update frequency and content quality through the Lyapunov optimization model, both caching efficiency and user experience can be improved.

[0111] The above descriptions are merely preferred embodiments of this application, and the present invention is not limited to the above embodiments. It is understood that other improvements and variations directly derived or conceived by those skilled in the art without departing from the spirit and concept of the present invention should be considered to be included within the protection scope of the present invention.

Claims

1. A collaborative optimization method for multi-layer edge caching networks, characterized in that, The method includes performing the following operations when a user terminal requests content from a small base station (SBS): On the macro base station MBS side, based on the content popularity distribution prediction results, the optimal update frequency and interval of each content file are determined in order to maximize the timeliness reward of long-term cached content files under the constraint of transmission cost. On the SBS side, the timeliness reward of short-term cached content files is calculated based on the global cache update decision and content popularity distribution prediction results obtained from the MBS side. Under the Lyapunov optimization framework, a virtual queue is introduced and the objective function of the drift plus penalty DPP term is used to realize the dynamic control of the local cache update decision of each time slot content file, so as to balance the cache update cost and the timeliness reward of the short-term cached content files. Based on the global cache update decision and the local cache update decision, the cache content on the MBS side and the SBS side are updated respectively in each time period to improve the timeliness of cache content in the overall network and respond to user content requests. The timeliness rewards for the MBS side and the SBS side are determined based on the information age (AoI) of the content and the user's perception function of the content's timeliness.

2. The multi-layer edge caching network collaborative optimization method according to claim 1, characterized in that, Introducing maximum information age As a content expiration threshold used to determine whether content is valid, the timeliness reward function constructed based on the AoI value and the perception function is defined as follows under different content expiration thresholds: Wherein, the perception function is a variant of the Sigmoid function; A For each content file, the AoI value in the corresponding content providing device is initialized to 1. When MBS or SBS updates the content file from the upper-level node, its corresponding AoI value is reset to the AoI of the source end; otherwise, it increases linearly over time.

3. The multi-layer edge caching network collaborative optimization method according to claim 1, characterized in that, The process of determining the optimal update frequency and interval for each content file based on the content popularity distribution prediction results, in order to maximize the timeliness reward of long-term cached content files under the constraint of transmission cost, includes: Based on the current MBS cycle interval Content popularity distribution prediction results The timeliness bonus for MBS cached content files is calculated as follows: in, Update content files for MBS The updated decision variables are based on the given... Calculate the time period The average timeliness bonus within is: in, For each content file on the MBS side f Time-sensitive rewards; Time With update interval Divide the data into segments, calculate the time-sensitive rewards for each update interval, and accumulate the time. If the time-sensitive reward is considered over all time intervals, then the MBS caching optimization problem is modeled as a function of update time intervals. Optimization problem; Determine the time The contents of the file Update count Then, combined with the predicted MBS content popularity distribution The MBS caching optimization problem is further decomposed into independent subproblems for each content file, to solve for the optimal update interval of each content file. These subproblems are expressed as follows: In the solution process, the KKT conditions are used to obtain the optimal update interval for each content file, and the sequential quadratic programming algorithm is used to solve for the optimal update frequency for each content file.

4. The multi-layer edge caching network collaborative optimization method according to claim 3, characterized in that, The process of obtaining the optimal update interval for each content file using KKT conditions and solving for the optimal update frequency for each content file using a sequential quadratic programming algorithm includes: Since the subproblem of the content file is a convex optimization problem, the KKT conditional solution is used to obtain the content file. The optimal update interval is and the update frequency is This indicates that, with a limited number of updates, maintaining the same update interval is the optimal strategy. Substituting the update frequency into the equation yields different content expiration thresholds. The average timeliness bonus is: After obtaining the optimal update interval, the MBS cache optimization problem is transformed into solving for the update frequency of each content file, that is, using a sequential quadratic programming algorithm to solve for the optimal update frequency of each content file. The content file cache update decision is defined as an update frequency vector. The optimization objective is to maximize the average timeliness reward of cached content files, expressed as: in, The total update frequency constraint for MBS to obtain content files from the previous node device within each time slot.

5. The multi-layer edge caching network collaborative optimization method according to claim 1, characterized in that, The Lyapunov optimization framework introduces a virtual queue and uses a drift-penalized DPP term as the objective function to dynamically control the local cache update decisions of content files in each time slot, balancing the cache update cost and the timeliness reward of the short-term cached content files. This includes: By introducing a virtual queue to constrain the cache update frequency, the following optimization framework is constructed to solve the SBS content file update decision variables. : in, For the SBS side of the first u Each node in the time slot The content file obtained inside Weighted time-sensitive rewards; It is the upper limit of the average update cost per time slot. It is the updated decision variable; , These are the AoI values ​​of each content file on the MBS side and the SBS side, respectively. Introducing the following Lyapunov virtual queue To measure the cost pressure of SBS cache updates: in, This indicates that SBS is in the time slot. The total number of updated content files; the growth of this virtual queue indicates that cache updates are too frequent and may exceed the system's update cost budget, while queue convergence indicates that update costs are effectively controlled; To dynamically adjust the cache update strategy, the drift plus penalty DPP term from Lyapunov optimization is adopted, denoted as: in, Indicates Lyapunov drift. This is a Lyapunov function used to measure the severity of queue backlog. For the system's immediate penalty function, corresponding to negative immediate rewards, It is an adjustable control parameter used to balance the relationship between time-sensitive rewards and queue stability; With the goal of maximizing system stability and timeliness rewards, in each time slot Minimize the DPP term to obtain the SBS content file update decision variable.

6. The multi-layer edge caching network collaborative optimization method according to claim 5, characterized in that, In each time slot The DPP terms to be minimized above include: The upper bound of the drift of the DPP term is established as follows: Let constant Then the DPP term simplifies to: in This represents the expected time-sensitive reward; ultimately, SBS updates the decision variables by solving the following minimization problem in each time slot: 。 7. The multi-layer edge caching network collaborative optimization method according to claim 1, characterized in that, The timeliness reward for short-term cached content files is calculated based on the global cache update decision and content popularity distribution prediction results obtained from the MBS side, including: The SBS side in the time slot The timeliness reward for content files obtained within the period is represented as follows: in, Indicates in time slot The first SBS side mentioned inside u Each node receives a content file. The number of requests, T( For each content file on the SBS side f Time-sensitive rewards; To measure in the current time slot Update content files The resulting timeliness benefit is redefined as the difference between the timeliness reward of the content file on the MBS side and the SBS side; based on the current SBS time slot. Content popularity distribution prediction results Further calculation of content files In the time slot The weighted time-sensitive reward within is: in, Each content file on the MBS side f Time-sensitive rewards It is an SBS-side content file. The predicted popularity value.

8. The multi-layer edge caching network collaborative optimization method according to claim 1, characterized in that, The methods for obtaining the content popularity distribution prediction results include: Content popularity distribution prediction is implemented using an LSTM neural network model to capture the temporal correlation of content request sequences. The LSTM model consists of two LSTM layers and a fully connected output layer, and uses the Adam optimizer for parameter updates, with mean squared error (MSE) as the loss function for training. The prediction process includes: On the SBS side, at the beginning of each cache update cycle, the content request data from the previous cycle is statistically analyzed into a local popularity distribution. The data is then fed into a trained LSTM model, which outputs the predicted content popularity distribution for the current period. It is used to guide SBS's current cache update decisions; On the MBS side, during each long-term period interval Within, historical content requests will be distributed. The input is fed into a trained LSTM model, which outputs the predicted content popularity distribution for the current period. It is used to guide MBS content collection optimization decisions.

9. A multi-layer edge caching network collaborative optimization system, characterized in that, The system includes multiple content providing devices, macro base stations (MBS), and multiple small base stations (SBS), wherein: The MBS is located at the center of the network and is used to collect content files generated by the content providing devices within the area and distribute the content to the SBS within the area for caching and storage. SBS nodes are evenly deployed within the coverage area of ​​the MBS, periodically receive content updates from the MBS, and provide real-time caching services to surrounding user terminals through caching strategies; The MBS and the SBS are used to collaboratively execute the method as described in any one of claims 1 to 8 to achieve collaborative scheduling of global and local cache update decision optimization in a multi-layer edge caching network.

Images