Data cache optimization method and device, equipment and storage medium

Abstract

The invention discloses a data cache optimization method and device, equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring a data set about request data of users corresponding to a plurality of edge servers, wherein the data set comprises a user request content sequence, network state information, cache content information and content popularity; inputting the data set corresponding to each edge server into a pre-trained decision model deployed in each edge server, and generating a caching strategy of each edge server; the cache strategy comprises one or more of cache replacement operation, content prefetching operation or pushing operation; and under the condition of executing the pushing operation, determining a pushing strategy for pushing the interested content of the user to the user according to the interested content of the user, the network state information and the cached content information of the edge server corresponding to the user, which are obtained by predicting the user request content sequence. According to the method, the user data privacy is effectively protected, and the communication overhead and the system delay are reduced.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data caching optimization method, apparatus, device, and storage medium. Background Technology

[0002] With the widespread adoption of mobile internet, IoT, and 5G technologies, multimedia data traffic generated by network edge devices has exploded. To alleviate the pressure on backhaul links and reduce user access latency, edge caching technology has emerged. This technology pre-caches popular multimedia content to network edge nodes (such as base stations and edge servers) closer to users, thereby enabling rapid local responses to user requests.

[0003] However, existing technologies typically employ a centralized AI training approach, where user data from all edge nodes is collected in a central server for unified model training and decision generation. Ultimately, the central server distributes the trained model to each edge node for execution. While this centralized approach can optimize caching strategies using global data, it is prone to data privacy leaks due to the need to transmit and store large amounts of raw user data, and it introduces high communication latency, thereby affecting the real-time performance and efficiency of cache updates. Summary of the Invention

[0004] This application provides a data caching optimization method, apparatus, device, and storage medium, which solves the problem that existing data caching optimization schemes, which rely on collecting user data from all edge nodes and sending it to a central server for unified model training and decision generation, are prone to data privacy leaks and introduce high communication latency. This effectively protects user data privacy, significantly reduces communication overhead and system latency, and improves the real-time performance and efficiency of cache updates.

[0005] Firstly, this application provides a data caching optimization method, including: Obtain a dataset of request data for users corresponding to multiple edge servers, wherein the dataset includes user request content sequence, network status information, cached content information, and content popularity; The datasets corresponding to each edge server are input into the pre-trained decision models deployed on each edge server to generate caching strategies for each edge server; wherein, the caching strategy includes one or more of cache replacement operations, content prefetching operations, or push operations; When the push operation is performed, a push strategy is determined to push the user's interested content to the user based on the user's requested content sequence predicted by the user, the network status information, and the cached content information of the edge server corresponding to the user.

[0006] Optionally, the trained decision model includes: The preset decision models deployed on each edge server are trained based on the user data corresponding to each edge server to obtain the first model parameters of each preset decision model. By aggregating the first model parameters of each of the preset decision models, the second model parameters are obtained; Based on the second model parameters, the first model parameters of each edge server corresponding to the preset decision model are updated to obtain the trained decision model. Optionally, the step of training a preset decision model deployed on each edge server based on user data corresponding to each edge server to obtain the first model parameters of each preset decision model includes: The dataset corresponding to each edge server is used as the state input to the preset decision model deployed on each edge server. The preset decision model generates the probability distributions corresponding to the execution of the cache replacement operation, the content prefetching operation and the push operation, respectively. Based on the probability distribution, select to perform at least one of the following operations: the cache replacement operation, the content prefetching operation, and the push operation; The metrics of each edge server after performing the corresponding operation are obtained. Based on the metrics and a preset multi-objective reward function, the reward value for performing the operation is calculated. With the goal of maximizing the accumulated reward value, the parameters of each preset decision model are updated to obtain the first model parameters of each preset decision model.

[0007] Optionally, the aggregation of the first model parameters of each of the preset decision models to obtain the second model parameters includes: The weight of each dataset corresponding to each edge server is determined based on the size of the dataset corresponding to each edge server; wherein, the dataset weight is expressed as the ratio of the dataset size of each edge server to the total dataset size of all edge servers. The first model parameters of the preset decision model corresponding to each edge server are weighted and averaged according to the dataset weights to obtain the second model parameters.

[0008] Optionally, predicting the content of interest to the user based on the user-requested content sequence includes: Obtain temporal features characterizing user behavior patterns from the sequence of user request content; Based on the aforementioned temporal features, local features for characterizing short-term user interests and global features for characterizing long-term user interests are determined. The local features and the global features are fused to obtain the fused features; Based on the fusion features, the user's interest rating for each piece of content is determined, and the content the user is interested in is predicted based on the interest rating.

[0009] Optionally, determining a push strategy for pushing the user's interested content to the user based on the user request content sequence predicted from the user's content sequence, the network status information, and the cached content information of the edge server corresponding to the user includes: Determine whether the content the user is interested in is cached on the edge server corresponding to the user; If the content of interest to the user is cached on the edge server corresponding to the user, a first push bitrate is determined based on the current network status information, and the content of interest to the user cached on the edge server corresponding to the user is pushed to the user at the first push bitrate. If the content of interest to the user is not cached on the edge server corresponding to the user, then the caching status of the content of interest to the user is queried from the neighboring edge server; When the user's interest content is obtained from the nearby edge server, the user's interest content is cached in the edge server corresponding to the user, and a second push bitrate is determined according to the current network status information. The user's interest content cached in the edge server corresponding to the user is then pushed to the user at the second push bitrate.

[0010] Optional, also includes: When performing the cache replacement operation, the set of cached content in the cache is determined based on the current cache content information; Based on the content popularity of each cached content in the cached content set, determine the content to be replaced from each cached content, remove the content to be replaced from the cache of the corresponding edge server, and cache the new content to the corresponding edge server; When performing the content prefetching operation, the content that the user may request in the future based on the user request sequence will be used as the prefetched content. Check whether the prefetched content is already cached on the edge server corresponding to the user. If the prefetched content is not cached on the edge server corresponding to the user, obtain the prefetched content from a neighboring edge server and cache it on the edge server corresponding to the user.

[0011] Secondly, this application provides a data caching optimization apparatus, comprising: The data acquisition module is used to acquire a dataset of request data for users corresponding to multiple edge servers. The dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The strategy generation module is used to input the dataset corresponding to each edge server into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server; wherein, the caching strategy includes one or more of cache replacement operation, content prefetching operation, or push operation; The push strategy determination module is used to determine a push strategy for pushing the user's interested content to the user based on the user's requested content sequence predicted by the user, the network status information, and the cached content information of the edge server corresponding to the user when the push operation is performed.

[0012] Thirdly, this application provides a data caching optimization device, comprising: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the data caching optimization method as described in the first aspect.

[0013] Fourthly, this application provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the data caching optimization method as described in the first aspect.

[0014] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: In this application, a dataset of user request data corresponding to multiple edge servers is obtained. This dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The dataset corresponding to each edge server is input into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server. The caching strategy includes one or more of cache replacement, content prefetching, or push operations. When a push operation is executed, a push strategy is determined based on the user's predicted content of interest from the user request content sequence, network status information, and the cached content information of the user's corresponding edge server, to push the content of interest to the user. In this technical approach, each edge node locally collects user request data and generates a caching strategy using a locally pre-trained decision model, without uploading the original data to a central server. This avoids centralized storage and transmission of original user data, fundamentally eliminating the risk of data privacy leakage. Simultaneously, it eliminates the original data transmission link from edge node to central server, reducing communication traffic consumption and avoiding delays caused by waiting for unified decisions from the central server, significantly improving the real-time performance and efficiency of cache updates. In addition, based on network status information and cached content information of edge servers, a push strategy is determined to push content that users are interested in to the user. This can accurately push content that users may be interested in to the user's end at the most appropriate bitrate, ensuring the smoothness of the content received by the user. Attached Figure Description

[0015] Figure 1 This is a flowchart of a data caching optimization method provided in an embodiment of this application; Figure 2 This is a flowchart of the trained decision model provided in the embodiments of this application; Figure 3 This is a flowchart illustrating the prediction of content of interest to the user, provided in an embodiment of this application. Figure 4 This is a flowchart illustrating the push strategy for determining content that a user is interested in, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the structure of a data caching optimization device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a data caching optimization device provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. A process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. A process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0017] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0018] In existing technologies, a centralized artificial intelligence training method is usually adopted. That is, user data from all edge nodes are collected in a central server for unified model training and decision generation. Finally, the central server distributes the trained model to each edge node for execution. Although this centralized method can optimize caching strategies using global data, it is prone to data privacy leakage risks due to the need to transmit and store a large amount of raw user data, and introduces high communication latency, thereby affecting the real-time performance and efficiency of cache updates.

[0019] To address the issues of existing data caching optimization schemes that rely on collecting user data from all edge nodes and sending it to a central server for unified model training and decision generation, which can easily lead to data privacy leaks and introduce high communication latency, this application provides a data caching optimization method. This method collects user request data locally at each edge node and generates a caching strategy using a pre-trained local decision model, eliminating the need to upload raw data to a central server. This avoids centralized storage and transmission of raw user data, fundamentally eliminating the risk of data privacy leaks. Furthermore, it eliminates the raw data transmission link between edge nodes and the central server, reducing communication traffic consumption and avoiding delays caused by waiting for unified decisions from the central server, significantly improving the real-time performance and efficiency of cache updates. In addition, based on network status information and the cached content information of the edge servers, a push strategy is determined to deliver content of interest to the user. This ensures that content the user may be interested in is accurately pushed to the user at the most suitable bitrate, guaranteeing a smooth user experience.

[0020] The data caching optimization method provided in this embodiment can be executed by a data caching optimization device, which can be implemented by software and / or hardware. The data caching optimization device can be composed of two or more physical entities, or it can be composed of a single physical entity.

[0021] The data caching optimization device may have an operating system installed, including but not limited to Android, Linux, and Windows. The data caching optimization device may also have applications installed on top of the operating system; these applications may be built into the operating system or downloaded from third-party devices or servers. In this embodiment, the data caching optimization device has at least one application capable of executing data caching optimization methods.

[0022] The technical solutions provided by the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] Figure 1 A flowchart of a data caching optimization method provided in an embodiment of this application is given. Please refer to... Figure 1 The data caching optimization method specifically includes: S110. Obtain a dataset of request data for users corresponding to multiple edge servers, wherein the dataset includes user request content sequence, network status information, cached content information, and content popularity.

[0024] An edge server is a server device deployed at the edge of the network (close to the user terminal). Each edge server is responsible for providing services to users within its coverage area (e.g., a business district, a residential area, or a stadium).

[0025] User request content sequence refers to the time-series record of multimedia data / content initiated by users within the service range of the edge server to the edge server. It reflects the patterns of user request behavior, such as when and how often content is requested. Optionally, the user request content sequence may include user identifiers, timestamps, and the requested content. Network status information refers to the communication link status parameters between the edge server and user terminals within its service range. It is used to assess the current network transmission capacity. Network status information may include bandwidth, latency (round-trip time for the edge server to send data to the user terminal and receive and return a response), jitter (the fluctuation range of network latency, reflecting network stability), etc.

[0026] Cached content information refers to information about the currently cached content on the edge server, including the cached content, the size of each cached item, and the caching time. Content popularity refers to the frequency with which a certain piece of data is requested by users within the service range of the edge server within a preset statistical period (such as 12 hours or 24 hours), reflecting the popularity of the content.

[0027] S120. Input the dataset corresponding to each edge server into the pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server; wherein, the caching strategy includes one or more of cache replacement operation, content prefetching operation or push operation.

[0028] The pre-trained decision model in this application is a deep reinforcement learning model trained on a federated learning architecture and deployed locally on various edge servers. This model models the caching decision problem as a Markov Decision Process (MDP), learning the optimal caching policy based on the state, the actions to be performed, and the reward function. In one embodiment, the decision model employs a proximal policy optimization (PPO) algorithm and an Actor-Critic architecture to ensure training stability in a continuous action space. The Actor network is responsible for outputting an action probability distribution based on the input state, suggesting the likelihood of performing various operations (such as cache replacement, prefetching, and push). The Critic network is responsible for evaluating the expected long-term cumulative reward in the current state, which guides the updates of the Actor network.

[0029] For example, Figure 2 This is a flowchart illustrating the trained decision model provided in the embodiments of this application. For example... Figure 2 As shown, the specific steps for obtaining the trained decision model include S1201-S1203: S1201. Train the preset decision models deployed on each edge server based on the user data corresponding to each edge server to obtain the first model parameters of each preset decision model.

[0030] The first model parameter refers to the updated set of internal parameters obtained by the preset decision model of each edge server after completing one round (or one cycle) of local training. The parameter set includes the weights and biases in the neural network.

[0031] In one embodiment, a preset decision model deployed on each edge server is trained based on user data corresponding to each edge server to obtain the first model parameters of each preset decision model. The steps include: S12011. The dataset corresponding to each edge server is used as the state input to the preset decision model deployed on each edge server. The preset decision model generates the probability distributions corresponding to the execution of cache replacement operation, content prefetching operation and push operation, respectively.

[0032] Cache replacement refers to removing one or more existing contents from the cache to free up space. For example, when the cache space of the edge server is insufficient (e.g., occupancy rate ≥ threshold) or the access frequency of cached content is low or the popularity of the content is declining, some inefficient cached content is removed to free up space to store new content (such as content with high popularity and high user request probability), avoiding waste of cache resources. Content prefetching refers to pulling content that is predicted to be requested by the user from nearby edge servers and caching it to the user's local edge server before the user makes a request. In essence, it is to reserve the content that the user may need in advance, reducing the probability of cache miss when the user makes a request, which can reduce the return latency. Push operation can be understood as pushing content that the edge server predicts the user's interest to the user.

[0033] Specifically, the dataset corresponding to each edge server is used as the state space input to the preset decision model deployed on each edge server. The feature extraction layer of the Actor network of the preset decision model extracts the high-level features of the input state space. The high-level features are passed to the output layer, which calculates and outputs the probability distributions corresponding to cache replacement operation, content prefetching operation and push operation.

[0034] S12012. Based on the probability distribution, select and execute at least one of the following operations: cache replacement operation, content prefetching operation, and push operation.

[0035] In this embodiment of the application, the operation can be selected for execution by random sampling according to the probability distribution. For example, a random number p between 0 and 1 is first generated. If it is less than a preset value, an operation is randomly selected; otherwise, the operation with the highest probability is selected for execution.

[0036] S12013. Obtain the metrics after each edge server performs the corresponding operation. Based on the metrics and the preset multi-objective reward function, calculate the reward value of the operation to maximize the accumulated reward value. Update the parameters of each preset decision model to obtain the first model parameters of each preset decision model.

[0037] The metric refers to the objective result generated within a preset time period after the selected operation is performed on the edge server, serving as the factual basis for evaluating the quality of the selected operation. Optionally, this metric includes cache hit rate H, average latency L, and energy consumption E. The cache hit rate H refers to the percentage of user-initiated requests that can directly obtain services from the cached content of the user's corresponding edge server within the preset time period after the corresponding operation is performed, out of the total number of requests. Its calculation formula is as follows:

[0038] in, For binary decision variables, representing content Whether it is cached, if the content If cached, then =1, otherwise =0; N is the total number of contents; It is content The popularity probability is calculated based on the content popularity model (assuming popularity follows a Zipf distribution), and the formula is:

[0039] Average latency L refers to the average time required to process all user requests within a preset time period. Its calculation formula is:

[0040] in, This indicates the return delay when a miss occurs. For transmission delay, To address latency. The lower the average latency, the faster users receive content, and the higher the quality of the user experience.

[0041] Energy consumption E refers to the total energy consumed by the edge server in executing operations in the caching strategy and providing services within a preset time period. It includes cache storage energy consumption and transmission energy consumption. Storage energy consumption is proportional to the size of the cached content, while transmission energy consumption is calculated based on transmission power and time.

[0042] In this embodiment, the reward value for the performed operation is calculated based on the indicator and a preset multi-objective reward function. The calculation formula is as follows:

[0043] in, The weight corresponding to the cache hit rate H. The weights corresponding to the average delay L are: The weight corresponding to energy consumption E.

[0044] Furthermore, this application embodiment also defines a multi-objective optimization function for jointly optimizing cache hit rate, average latency, and energy consumption. The optimization problem is expressed as follows:

[0045] The constraints are:

[0046] in, Yes, it refers to the content size. It's the cache capacity. These are the cache hit rate functions. Delay function and energy consumption function The corresponding weights.

[0047] The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method, and solved using gradient descent. The optimal cached decision variable is found through iterative updates. The Lagrange function is:

[0048] S1202. Aggregate the first model parameters of each of the preset decision models to obtain the second model parameters.

[0049] In this embodiment, the central server periodically collects the first model parameters of the preset decision model deployed on each edge server and aggregates them to obtain the second model parameters.

[0050] In one embodiment of this application, the aggregation step includes: determining the dataset weights corresponding to each edge server based on the size of the datasets corresponding to each edge server; wherein, the dataset weight represents the ratio of the dataset size of each edge server to the total dataset size of all edge servers; and weighting the first model parameters of the preset decision model corresponding to each edge server according to the dataset weights to obtain the second model parameters. The aggregation formula is:

[0051] in, It is the number of edge nodes. This refers to the dataset size. This aggregation method ensures that the decision model adapts to the global data distribution while reducing communication overhead.

[0052] S1203. Based on the second model parameters, update the first model parameters of the preset decision model corresponding to each edge server to obtain the trained decision model. After obtaining the aggregated second model parameters, the central server will distribute the second model parameters to the corresponding preset models of each edge server.

[0053] In this embodiment, each edge node locally collects user request data and generates a caching strategy using its pre-trained decision model. The first model parameters of each model are aggregated only in the central server, without uploading the original dataset to the central server. This achieves distributed optimization through federated learning, avoiding centralized storage and transmission of the user's original dataset, fundamentally eliminating the risk of data privacy leakage. At the same time, it eliminates the original dataset transmission link from edge node to central server, reducing communication traffic consumption and avoiding delays caused by waiting for unified decision-making by the central server, significantly improving the real-time performance and efficiency of cache updates.

[0054] S130. When performing the push operation, a push strategy is determined to push the user's interested content to the user based on the user's requested content sequence predicted by the user, the network status information, and the cached content information of the edge server corresponding to the user.

[0055] Specifically, when the caching strategy output by the pre-trained decision model includes push operations, the push strategy needs to be determined first, and then the content that the user is interested in should be pushed to the user according to the determined push strategy. The purpose of the push strategy is to proactively deliver the content that the user is most likely to be interested in at the right time, with the best quality and efficiency, thereby improving the user experience and optimizing the utilization of network resources.

[0056] This application embodiment uses a Long Short-Term Memory (LSTM) network to predict the content a user is interested in based on a sequence of content requested by the user. For example, Figure 3 This is a flowchart illustrating the prediction of content of interest to the user, provided in an embodiment of this application. For example... Figure 3 As shown, the steps for predicting content of interest to the user specifically include S1301-S1304: S1301. Obtain temporal features representing user behavior patterns from the user request content sequence.

[0057] The temporal features of user behavior patterns refer to high-dimensional vector representations extracted from the sequence of user-requested content that quantify the dynamic access patterns of users. They reflect behavioral characteristics such as the periodicity of user access, content relevance, and interest shifting patterns by capturing the contextual dependencies and evolutionary trends of request behavior over time. Specifically, the sequence of user-requested content is first preprocessed, transforming discrete content identifiers into continuous embedding vectors to obtain a standardized sequence. This sequence is then input into an LSTM network. The LSTM network iteratively processes the input at each time step through its internal gating mechanisms (including forget gate, input gate, and output gate). The forget gate controls the degree of retention of historical information, the input gate determines the magnitude of new information updates, cell states act as memory units integrating temporal information, and the output gate generates the hidden states for each time step. After processing the complete sequence, the series of hidden states output by the LSTM network constitutes the temporal features.

[0058] S1302. Based on the time-series features, determine local features for characterizing short-term user interests and global features for characterizing long-term user interests.

[0059] For example, local features representing a user's short-term interests can be obtained by extracting the hidden states of the last few time steps of the LSTM network; these features represent changes in the user's recent requested content behavior. Alternatively, the final hidden states of the sequence can be directly used as global features, or the hidden states of all time steps in the entire sequence can be taken and weighted by averaging them according to the lower weights of earlier steps and the higher weights of more recent steps to obtain global features.

[0060] S1303. The local features and the global features are fused to obtain the fused features.

[0061] For example, the vectors of local features and global features are first concatenated to obtain the concatenated feature vector, which is then used to obtain the fused features.

[0062] S1304. Based on the fusion features, determine the user's interest rating for each piece of content, and predict the content the user is interested in based on the interest rating.

[0063] For example, the fused features are input into a fully connected layer, which performs a linear transformation on the fused features to calculate the original rating vector for each content. Then, a softmax function is performed on the original rating vector to normalize it, resulting in an interest score for each content. The higher the score, the stronger the user's interest.

[0064] For example, Figure 4 This is a flowchart illustrating the push strategy for determining content that a user is interested in, provided in an embodiment of this application. For example... Figure 4As shown, the steps for determining the push strategy of sending content that the user is interested in specifically include S1305-S1308: S1305. Determine whether the content that the user is interested in is cached in the edge server corresponding to the user.

[0065] S1306. If the content of interest to the user is cached on the edge server corresponding to the user, determine the first push bitrate based on the current network status information, and push the content of interest to the user cached on the edge server corresponding to the user to the user at the first push bitrate.

[0066] In this embodiment, the push bitrate refers to the amount of data sent per unit time when the edge server actively transmits content of interest to the user terminal, typically measured in Mbps (megabits per second). Its function is to balance the image quality and smoothness of content transmission. A higher bitrate results in clearer content details but requires higher network capacity (e.g., greater network bandwidth); a lower bitrate results in more basic image quality but requires less network bandwidth and stability, thus avoiding transmission stuttering. For example, after determining that the content of interest to the user has been cached on the edge server corresponding to the user, the network status information of the communication link between the user's local edge server and the target user terminal is obtained. The network status information may include bandwidth, jitter, etc. If the current bandwidth is greater than or equal to a preset bandwidth threshold and the current jitter is less than or equal to a preset jitter threshold, the current network status information is determined to be good, and the first bitrate is determined to be the high-definition bitrate (the high-definition bitrate provides higher resolution, with sufficient bandwidth and stable network, suitable for pushing 1080p high-definition content), and the content of interest to the user is pushed to the user at the high-definition bitrate; otherwise, the first push bitrate is determined to be the standard-definition bitrate (the standard-definition bitrate provides lower resolution, with limited bandwidth or greater network jitter, which can reduce the risk of transmission interruption), and the content of interest to the user is pushed to the user at the standard-definition bitrate.

[0067] S1307. If the content of interest to the user is not cached on the edge server corresponding to the user, then query the caching status of the content of interest to the user from the neighboring edge server.

[0068] Among them, a neighboring edge server refers to a neighboring edge server that meets the low-latency communication conditions in terms of physical location and network topology with respect to the edge server corresponding to the user (i.e., the edge server that initiated the query). S1308. When the user's interest content is obtained from the nearby edge server, the user's interest content is cached in the edge server corresponding to the user, and a second push bitrate is determined according to the current network status information. The user's interest content cached in the edge server corresponding to the user is then pushed to the user at the second push bitrate.

[0069] Specifically, after caching the content of interest to the user to the user's corresponding edge server, the network status information of the communication link between the user's corresponding edge server and the target user terminal is obtained. If the current bandwidth is greater than or equal to a preset bandwidth threshold and the current jitter is less than or equal to a preset jitter threshold, the current network status information is determined to be good, the second bitrate is determined to be the high-definition bitrate, and the content of interest to the user is pushed to the user at the high-definition bitrate; otherwise, the second push bitrate is determined to be the standard-definition bitrate, and the content of interest to the user is pushed to the user at the standard-definition bitrate.

[0070] The embodiments of this application determine the push bitrate based on the current network status information, aiming to provide users with the best possible and smoothest experience under the current network conditions. In one embodiment of this application, when the caching strategy output by the pre-trained decision model includes a cache replacement operation, when performing the cache replacement operation, the set of cached content in the cache is determined according to the current cached content information; the content to be replaced is determined from each cached content according to the content popularity of each cached content in the cached content set, and the content to be replaced is removed from the cache of the corresponding edge server, and the new content is cached to the corresponding edge server.

[0071] The content to be replaced refers to the content selected for removal from the cache according to a preset strategy when performing cache replacement operations on the edge server. Specifically, firstly, all cached content is sorted in ascending order of content popularity, with content having lower popularity being replaced first. Secondly, the size of the new content to be cached is determined (e.g., if the new content is 60MB), and the space to be freed from the cached content must be greater than or equal to the size of the new content (e.g., if the new content is 60MB, the space to be freed from the cached content must be ≥60MB). Cached content is then selected sequentially from the sorted results, and the size of the selected cached content is accumulated until its size is greater than or equal to the required freed space. The selected cached content is then designated as the content to be replaced.

[0072] In one embodiment of this application, when the caching strategy output by the pre-trained decision model includes a content prefetching operation, when performing the content prefetching operation, the content that the user may request in the future based on the user request sequence is used as the prefetched content; it is checked whether the prefetched content is already cached on the edge server corresponding to the user; if the prefetched content is not cached on the edge server corresponding to the user, the prefetched content is obtained from a neighboring edge server and cached on the edge server corresponding to the user.

[0073] In summary, the data caching optimization method provided in this application involves acquiring a dataset of user request data corresponding to multiple edge servers. This dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The dataset corresponding to each edge server is input into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server. The caching strategy includes one or more of cache replacement, content prefetching, or push operations. When performing a push operation, a push strategy is determined based on the user's predicted content of interest from the user request content sequence, network status information, and the cached content information of the user's corresponding edge server. In this approach, each edge node locally collects user request data and generates a caching strategy using a locally pre-trained decision model, eliminating the need to upload the original data to a central server. This avoids centralized storage and transmission of original user data, fundamentally eliminating the risk of data privacy leakage. Furthermore, it eliminates the original data transmission link between the edge node and the central server, reducing communication traffic consumption and avoiding delays caused by waiting for unified decisions from the central server, significantly improving the real-time performance and efficiency of cache updates. In addition, based on network status information and cached content information of edge servers, a push strategy is determined to push content that users are interested in to the user. This can accurately push content that users may be interested in to the user's end at the most appropriate bitrate, ensuring the smoothness of the content received by the user.

[0074] Figure 5 This is a schematic diagram of a data caching optimization device provided in an embodiment of this application. (Reference) Figure 5 The data caching optimization device provided in this embodiment specifically includes: a data acquisition module 21, a strategy generation module 22, and a push strategy determination module 23.

[0075] The data acquisition module 21 is used to acquire a dataset of request data for users corresponding to multiple edge servers, wherein the dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The strategy generation module 22 is used to input the dataset corresponding to each edge server into the decision model pre-trained on each edge server to generate a caching strategy for each edge server; wherein, the caching strategy includes one or more of cache replacement operation, content prefetching operation or push operation; The push strategy determination module 23 is used to determine a push strategy for pushing the user's interested content to the user based on the user's requested content sequence predicted by the user's content sequence, the network status information, and the cached content information of the edge server corresponding to the user when the push operation is performed.

[0076] Based on the above embodiments, the strategy generation module 22 includes a model training unit, which includes: The training subunit is used to train the preset decision models deployed on each edge server based on the user data corresponding to each edge server, so as to obtain the first model parameters of each preset decision model. The aggregation subunit is used to aggregate the first model parameters of each of the preset decision models to obtain the second model parameters; The parameter update subunit is used to update the first model parameters of the preset decision model corresponding to each edge server based on the second model parameters, so as to obtain the trained decision model. Based on the above embodiments, the training subunit includes: The dataset corresponding to each edge server is used as the state input to the preset decision model deployed on each edge server. The preset decision model generates the probability distributions corresponding to the execution of the cache replacement operation, the content prefetching operation and the push operation, respectively. Based on the probability distribution, select to perform at least one of the following operations: the cache replacement operation, the content prefetching operation, and the push operation; The metrics of each edge server after performing the corresponding operation are obtained. Based on the metrics and a preset multi-objective reward function, the reward value for performing the operation is calculated. With the goal of maximizing the accumulated reward value, the parameters of each preset decision model are updated to obtain the first model parameters of each preset decision model.

[0077] Based on the above embodiments, the aggregation subunit includes: The weight of each dataset corresponding to each edge server is determined based on the size of the dataset corresponding to each edge server; wherein, the dataset weight is expressed as the ratio of the dataset size of each edge server to the total dataset size of all edge servers. The first model parameters of the preset decision model corresponding to each edge server are weighted and averaged according to the dataset weights to obtain the second model parameters.

[0078] Based on the above embodiments, the push strategy determination module 23 includes an interest content prediction unit, which includes: The temporal feature acquisition subunit is used to acquire temporal features representing user behavior patterns from the user request content sequence; The content feature determination subunit is used to determine, based on the temporal features, local features for characterizing short-term user interests and global features for characterizing long-term user interests. A feature fusion subunit is used to fuse the local features and the global features to obtain fused features; The Interest Content Prediction Subunit is used to determine the user's interest rating for each piece of content based on the fusion features, and to predict the content that the user is interested in based on the interest rating.

[0079] Based on the above embodiments, the push strategy determination module 23 includes a push strategy determination unit, which includes: The judgment subunit is used to determine whether the content that the user is interested in is cached on the edge server corresponding to the user; The first push strategy determination subunit is used to determine a first push bitrate based on the current network status information when the content of interest to the user is cached on the edge server corresponding to the user, and push the content of interest to the user cached on the edge server corresponding to the user to the user at the first push bitrate. The Interested Content Query Subunit is used to query the caching status of the user's interested content from a nearby edge server if the user's interested content is not cached on the edge server corresponding to the user. The second push strategy determination subunit is used to cache the user's interest content in the edge server corresponding to the user when the user's interest content is obtained from the nearby edge server, determine the second push bitrate according to the current network status information, and push the user's interest content cached in the edge server corresponding to the user to the user at the second push bitrate.

[0080] Based on the above embodiments, it also includes: When performing the cache replacement operation, the set of cached content in the cache is determined based on the current cache content information; Based on the content popularity of each cached content in the cached content set, determine the content to be replaced from each cached content, remove the content to be replaced from the cache of the corresponding edge server, and cache the new content to the corresponding edge server; When performing the content prefetching operation, the content that the user may request in the future based on the user request sequence will be used as the prefetched content. Check whether the prefetched content is already cached on the edge server corresponding to the user. If the prefetched content is not cached on the edge server corresponding to the user, obtain the prefetched content from a neighboring edge server and cache it on the edge server corresponding to the user.

[0081] In summary, the data caching optimization apparatus provided in this application embodiment acquires a dataset of request data for users corresponding to multiple edge servers. This dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The dataset corresponding to each edge server is input into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server. The caching strategy includes one or more of cache replacement, content prefetching, or push operations. When a push operation is executed, a push strategy is determined based on the user's predicted content of interest from the user request content sequence, network status information, and the cached content information of the edge server corresponding to the user, to push the content of interest to the user. In this technical approach, each edge node locally collects user request data and generates a caching strategy using a locally pre-trained decision model, without uploading the original data to a central server. This avoids centralized storage and transmission of original user data, fundamentally eliminating the risk of data privacy leakage. Simultaneously, it eliminates the original data transmission link from edge node to central server, reducing communication traffic consumption and avoiding delays caused by waiting for unified decisions from the central server, significantly improving the real-time performance and efficiency of cache updates. In addition, based on network status information and cached content information of edge servers, a push strategy is determined to push content that users are interested in to the user. This can accurately push content that users may be interested in to the user's end at the most appropriate bitrate, ensuring the smoothness of the content received by the user.

[0082] The data caching optimization device provided in this application embodiment can be used to execute the data caching optimization method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0083] Figure 6 This is a schematic diagram of the structure of a data caching optimization device provided in an embodiment of this application, with reference to... Figure 6The data caching optimization device includes a processor 31, a memory 32, a communication device 33, an input device 34, and an output device 35. The number of processors 31 and the number of memories 32 in the data caching optimization device can be one or more. The processor 31, memory 32, communication device 33, input device 34, and output device 35 of the data caching optimization device can be connected via a bus or other means.

[0084] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the data caching optimization method in any embodiment of this application (e.g., the data acquisition module 21, strategy generation module 22, and push strategy determination module 23 in the data caching optimization device). The memory 32 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on device usage, etc. Furthermore, the memory 32 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0085] The communication device 33 is used for data transmission.

[0086] The processor 31 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 32, thereby realizing the above-mentioned data cache optimization method.

[0087] Input device 34 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 35 may include display devices such as a display screen.

[0088] The data caching optimization device provided above can be used to execute the data caching optimization method provided in the above embodiments, and has corresponding functions and beneficial effects.

[0089] This application embodiment also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a data caching optimization method. The data caching optimization method includes: acquiring a dataset of requested data for users corresponding to multiple edge servers, wherein the dataset includes a sequence of user request content, network status information, cached content information, and content popularity; inputting the dataset corresponding to each edge server into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server; wherein the caching strategy includes one or more of cache replacement operations, content prefetching operations, or push operations; and, in the case of performing a push operation, determining a push strategy to push the user's interested content to the user based on the user's requested content sequence predicted from the user's content sequence, network status information, and cached content information of the edge server corresponding to the user.

[0090] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer system in which the program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term “storage medium” can include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0091] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the data cache optimization method described above, but can also perform related operations in the data cache optimization method provided in any embodiment of this application.

[0092] The data caching optimization apparatus, storage medium, and data caching optimization device provided in the above embodiments can execute the data caching optimization method provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the data caching optimization method provided in any embodiment of this application.

[0093] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application. The scope of this application is determined by the scope of the claims.

Claims

1. A data caching optimization method, characterized in that, include: Obtain a dataset of request data for users corresponding to multiple edge servers, wherein the dataset includes user request content sequence, network status information, cached content information, and content popularity; The datasets corresponding to each edge server are input into the pre-trained decision models deployed on each edge server to generate caching strategies for each edge server; wherein, the caching strategy includes one or more of cache replacement operations, content prefetching operations, or push operations; When the push operation is performed, a push strategy is determined to push the user's interested content to the user based on the user's requested content sequence predicted by the user, the network status information, and the cached content information of the edge server corresponding to the user.

2. The data caching optimization method according to claim 1, characterized in that, The trained decision model is obtained by training, including: The preset decision models deployed on each edge server are trained based on the user data corresponding to each edge server to obtain the first model parameters of each preset decision model. By aggregating the first model parameters of each of the preset decision models, the second model parameters are obtained; Based on the second model parameters, the first model parameters of each edge server corresponding to the preset decision model are updated to obtain the trained decision model.

3. The data caching optimization method according to claim 2, characterized in that, The process of training a pre-defined decision model deployed on each edge server based on user data corresponding to each edge server to obtain the first model parameters of each pre-defined decision model includes: The dataset corresponding to each edge server is used as the state input to the preset decision model deployed on each edge server. The preset decision model generates the probability distributions corresponding to the execution of the cache replacement operation, the content prefetching operation and the push operation, respectively. Based on the probability distribution, select to perform at least one of the following operations: the cache replacement operation, the content prefetching operation, and the push operation; The metrics of each edge server after performing the corresponding operation are obtained. Based on the metrics and a preset multi-objective reward function, the reward value for performing the operation is calculated. With the goal of maximizing the accumulated reward value, the parameters of each preset decision model are updated to obtain the first model parameters of each preset decision model.

4. The data caching optimization method according to claim 2, characterized in that, The aggregation of the first model parameters of each of the preset decision models to obtain the second model parameters includes: The weight of each dataset corresponding to each edge server is determined based on the size of the dataset corresponding to each edge server; wherein, the dataset weight is expressed as the ratio of the dataset size of each edge server to the total dataset size of all edge servers. The first model parameters of the preset decision model corresponding to each edge server are weighted and averaged according to the dataset weights to obtain the second model parameters.

5. The data caching optimization method according to claim 1, characterized in that, Based on the user request content sequence, the user's interested content is predicted, including: Obtain temporal features characterizing user behavior patterns from the sequence of user request content; Based on the aforementioned temporal features, local features for characterizing short-term user interests and global features for characterizing long-term user interests are determined. The local features and the global features are fused to obtain the fused features; Based on the fusion features, the user's interest rating for each piece of content is determined, and the content the user is interested in is predicted based on the interest rating.

6. The data caching optimization method according to claim 1, characterized in that, The step of determining a push strategy to push the user's interested content to the user based on the user request content sequence predicted from the network status information and the cached content information of the edge server corresponding to the user includes: Determine whether the content the user is interested in is cached on the edge server corresponding to the user; If the content of interest to the user is cached on the edge server corresponding to the user, a first push bitrate is determined based on the current network status information, and the content of interest to the user cached on the edge server corresponding to the user is pushed to the user at the first push bitrate. If the content of interest to the user is not cached on the edge server corresponding to the user, then the caching status of the content of interest to the user is queried from the neighboring edge server; When the user's interest content is obtained from the nearby edge server, the user's interest content is cached in the edge server corresponding to the user, and a second push bitrate is determined according to the current network status information. The user's interest content cached in the edge server corresponding to the user is then pushed to the user at the second push bitrate.

7. The data caching optimization method according to claim 1, characterized in that, Also includes: When performing the cache replacement operation, the set of cached content in the cache is determined based on the current cache content information; Based on the content popularity of each cached content in the cached content set, determine the content to be replaced from each cached content, remove the content to be replaced from the cache of the corresponding edge server, and cache the new content to the corresponding edge server; When performing the content prefetching operation, the content that the user may request in the future based on the user request sequence will be used as the prefetched content. Check whether the prefetched content is already cached on the edge server corresponding to the user. If the prefetched content is not cached on the edge server corresponding to the user, obtain the prefetched content from a neighboring edge server and cache it on the edge server corresponding to the user.

8. A data caching optimization device, characterized in that, include: The data acquisition module is used to acquire a dataset of request data for users corresponding to multiple edge servers. The dataset includes a sequence of user request content, network status information, cached content information, and content popularity. The strategy generation module is used to input the dataset corresponding to each edge server into a pre-trained decision model deployed on each edge server to generate a caching strategy for each edge server; wherein, the caching strategy includes one or more of cache replacement operation, content prefetching operation, or push operation; The push strategy determination module is used to determine a push strategy for pushing the user's interested content to the user based on the user's requested content sequence predicted by the user, the network status information, and the cached content information of the edge server corresponding to the user when the push operation is performed.

9. A data caching optimization device, characterized in that, include: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the data caching optimization method as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the data caching optimization method as described in any one of claims 1-7.

Images