CN118316890A - Content acceleration method, device, equipment and storage medium based on edge cache - Google Patents

Abstract

The present application relates to the field of data processing technology, and discloses a content acceleration method, device, equipment and storage medium based on edge cache. The method includes: collecting historical network traffic data, and extracting traffic characteristics of the historical network traffic data to obtain traffic characteristic data; constructing an edge node resource allocation strategy through traffic characteristic data, and generating a cache space priority table according to the edge node resource allocation strategy; collecting user data of the target user, and matching the user data with edge nodes through the edge node resource allocation strategy to obtain multiple initial edge nodes; based on the cache space priority table, constructing a node layout for multiple initial edge nodes to obtain a target node layout; obtaining the data access request of the target user, and matching the data access request with access nodes based on the target node layout to obtain the target access node. The present application improves the acceleration efficiency of content acceleration based on edge cache.

Description

Content acceleration method, device, equipment and storage medium based on edge cache

Technical Field

The present application relates to the field of data processing, and in particular to a content acceleration method, device, equipment and storage medium based on edge caching.

Background technique

In today's Internet era, content transmission speed and quality have become key indicators for measuring user experience. With the rise of online video streaming, social media, and cloud services, users' requirements for data transmission speed are increasing. To meet these needs, edge computing has emerged. By transferring data processing and storage from central servers to the edge of the network closer to users, it greatly reduces the delay of data transmission and improves the response speed of services. However, although edge computing can theoretically provide faster content access speed, in practical applications, how to efficiently manage edge nodes, intelligently schedule network resources, and how to dynamically optimize content distribution strategies according to network conditions are still major challenges.

Existing edge computing solutions have certain limitations when dealing with dynamic network environments and user needs. First, most solutions adopt static data caching and distribution strategies, which are difficult to adapt to the rapid changes in network status and the diversity of user access patterns. In addition, existing methods often lack sufficient flexibility and intelligence in resource allocation and load balancing, and cannot fully utilize the potential of edge computing to optimize content delivery. For example, when multiple users access popular content at the same time, static caching strategies may cause some edge nodes to be overloaded while other nodes have idle resources, thus affecting the overall service quality and user experience. Therefore, how to intelligently adjust the resource allocation and content distribution strategies of edge nodes according to real-time network conditions and user behavior has become the key to improving edge computing efficiency and optimizing user experience.

Summary of the invention

The present application provides a content acceleration method, apparatus, device and storage medium based on edge cache, which are used to improve the acceleration efficiency of content acceleration based on edge cache.

In a first aspect, the present application provides a content acceleration method based on edge cache, and the content acceleration method based on edge cache includes: collecting historical network traffic data, and extracting traffic features of the historical network traffic data to obtain traffic feature data; constructing an edge node resource allocation strategy through the traffic feature data, and generating a cache space priority table according to the edge node resource allocation strategy; collecting user data of a target user, and performing edge node matching on the user data through the edge node resource allocation strategy to obtain multiple initial edge nodes; based on the cache space priority table, performing node layout construction on the multiple initial edge nodes to obtain a target node layout; obtaining a data access request of the target user, and performing access node matching on the data access request based on the target node layout to obtain a target access node.

In combination with the first aspect, in a first implementation method of the first aspect of the present application, the historical network traffic data is collected and traffic feature extraction is performed on the historical network traffic data to obtain traffic feature data, including: collecting the historical network traffic data and performing information traversal on the historical network traffic data to obtain historical access frequency, historical access time and historical access content type; performing vector conversion on the historical access frequency to obtain a first vector, performing vector conversion on the historical access time to obtain a second vector, and at the same time, performing vector conversion on the historical access content type to obtain a third vector; performing vector fusion on the first vector and the second vector to obtain a first fused vector; performing vector fusion on the first vector and the third vector to obtain a second fused vector; performing vector fusion on the second vector and the third vector to obtain a third fused vector; inputting the first fused vector into a preset convolutional long short-term memory network model for Extract high-frequency access time features to obtain high-frequency access time features; input the second fusion vector into the convolutional long short-term memory network model to extract high-frequency access content features to obtain high-frequency access content features; input the third fusion vector into the convolutional long short-term memory network model to extract long-term access content to obtain long-term access content features; perform feature fusion on the high-frequency access time features and the high-frequency access content features to obtain a first fusion feature, and at the same time, perform feature fusion on the high-frequency access content features and the long-term access content features to obtain a second fusion feature; input the first fusion feature into the feature recognition layer of the convolutional long short-term memory network model to perform traffic pattern recognition to obtain a traffic pattern, and at the same time, input the second fusion feature into the feature recognition layer of the convolutional long short-term memory network model to perform content trend recognition to obtain an access content trend; merge the traffic pattern and the access content trend into the traffic feature data.

In combination with the first aspect, in a second implementation method of the first aspect of the present application, the edge node resource allocation strategy is constructed through the traffic feature data, and a cache space priority table is generated according to the edge node resource allocation strategy, including: performing pattern recognition on the traffic pattern to obtain pattern recognition data, wherein the pattern recognition data includes: periodic traffic change values and traffic peak values; predicting the bandwidth demand for the periodic traffic change values and the traffic peak values through a timing prediction algorithm to obtain bandwidth demand for multiple time periods; identifying peak time periods for bandwidth demand for multiple time periods to obtain traffic peak time periods; identifying valley time periods for bandwidth demand for multiple time periods to obtain traffic valley time periods; identifying high-demand content types for the access content trend to obtain a high-demand content type set; generating an initial resource allocation strategy based on the traffic peak time period and the high-demand content type set; performing policy modification on the initial resource allocation strategy through the traffic valley time period to obtain the edge node resource allocation strategy; prioritizing the high-demand content type set through the edge node resource allocation strategy to obtain sorting data, and generating a cache space priority table through the sorting data.

In combination with the first aspect, in a third implementation method of the first aspect of the present application, the user data of the target user is collected, and the user data is matched with edge nodes through the edge node resource allocation strategy to obtain multiple initial edge nodes, including: collecting the user data of the target user, wherein the user data includes user geographic location data and user expected access content data; performing server computing area identification on the user geographic location data to obtain a server computing area; performing data collection on edge node information in the server computing area to obtain regional node data; performing data fusion on the user geographic location data and the user expected access content data to obtain a user demand fingerprint; performing node resource traversal on the regional node data to obtain node computing power resources corresponding to multiple regional nodes, and at the same time, collecting resource load data of multiple regional nodes; performing node screening on the resource load data of multiple regional nodes through the user demand fingerprint to obtain multiple initial edge nodes.

In combination with the first aspect, in a fourth implementation method of the first aspect of the present application, the data access request of the target user is obtained, and access node matching is performed on the data access request based on the target node layout to obtain the target access node, including: obtaining the data access request of the target user, and parsing the data access request to obtain the data request frequency and the current server load data; obtaining the field of view data of the target user, and collecting the popularity value of the video content to be accessed, and inputting the field of view data and the popularity value set into a preset deep deterministic policy gradient algorithm to predict the focus area of the video content to obtain the user focus area; extracting the content to be cached for the video content to be accessed based on the user focus area to obtain the content to be cached; data caching is performed on the content to be cached through the target node layout, and based on the target node layout, the data request frequency and the current server load data are input into a preset dual deep Q network algorithm to match the access nodes to obtain the target access node.

In combination with the first aspect, in a fifth implementation of the first aspect of the present application, the field of view data of the target user is obtained, and the popularity value of the video content to be accessed is collected, and the field of view data and the popularity value set are input into a preset deep deterministic policy gradient algorithm to predict the focus area of the video content to obtain the user focus area, including: obtaining the field of view data of the target user, and extracting the video views of the video content to be accessed to obtain the video views; extracting the video likes of the video content to be accessed to obtain the video likes; extracting the video complete viewing amount of the video content to be accessed to obtain the video complete viewing amount; calculating the video popularity value of the video views, the video likes and the video complete viewing amount through a preset weight parameter set to obtain the popularity value set; calculating the video frame ratio value of the field of view data to obtain the video frame ratio value corresponding to the field of view data;

Based on the video frame ratio value, the field of view data and the popularity value set are input into the deep deterministic policy gradient algorithm, and the coordinates of the center point of the focus area are calculated by the center point coordinate calculation formula to obtain the coordinates of the center point of the focus area, wherein the center point coordinate calculation formula is as follows:

;

Where n is the number of data points of the field of view data, is the coordinate of the center point of the focal area, is the coordinate of the i-th field of view data point, is the angular offset of the i-th field of view data point relative to the center of the video content, is the first exponential parameter, is the second exponential parameter, is the weight factor, where The calculation formula is as follows;

;

in, is the popularity value corresponding to the i-th viewing angle data point, is the Euclidean distance from the i-th field of view data point to the center of the video content, is the correction factor;

Based on the coordinates of the center point of the focus area, multiple candidate focus areas are generated, and a reward score is calculated for each of the candidate focus areas through a preset reward function to obtain reward score data for each candidate focus area; based on the reward score data for each candidate focus area, multiple candidate focus areas are area screened to obtain the user focus area.

In combination with the first aspect, in a sixth implementation of the first aspect of the present application, data caching is performed on the content to be cached through the target node layout, and based on the target node layout, the data request frequency and the current server load data are input into a preset dual deep Q network algorithm for access node matching to obtain a target access node, including: performing a network connection status analysis on the target node layout to obtain a current network connection status; matching a data caching strategy based on the network connection status, and caching data on the content to be cached through the data caching strategy, wherein the data caching strategy includes a data allocation sub-strategy and a data transmission sub-strategy; inputting the data request frequency into the dual deep Q network The algorithm constructs a frequency state vector to obtain a multidimensional frequency state vector; the current server load data is input into the dual deep Q network algorithm to construct a load state vector to obtain a multidimensional load state vector; the multidimensional frequency state vector and the multidimensional load state vector are input into the main network of the dual deep Q network algorithm to perform action execution to obtain action execution state data and reward parameters; the action execution state data and the reward parameters are input into the target network of the dual deep Q network algorithm to perform maximum Q value dynamic analysis to obtain a maximum Q value action; the target Q value of the maximum Q value action is extracted, and the target node layout is matched with the access node based on the target Q value to obtain the target access node.

In a second aspect, the present application provides a content acceleration device based on edge cache, and the content acceleration device based on edge cache includes:

A collection module is used to collect historical network traffic data and extract traffic characteristics from the historical network traffic data to obtain traffic characteristic data;

A generation module, used to construct an edge node resource allocation strategy based on the traffic characteristic data, and generate a cache space priority table according to the edge node resource allocation strategy;

A matching module, used to collect user data of a target user, and perform edge node matching on the user data through the edge node resource allocation strategy to obtain a plurality of initial edge nodes;

A construction module, configured to construct a node layout for the plurality of initial edge nodes based on the cache space priority table to obtain a target node layout;

The acquisition module is used to acquire the data access request of the target user, and perform access node matching on the data access request based on the target node layout to obtain the target access node.

A third aspect of the present application provides an edge cache-based content acceleration device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the edge cache-based content acceleration device executes the above-mentioned edge cache-based content acceleration method.

A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned edge cache-based content acceleration method.

In the technical solution provided by this application, the efficiency and quality of content transmission are significantly improved through resource allocation and data transmission strategies. First, by collecting and analyzing historical network traffic data, the constructed edge node resource allocation strategy can dynamically adjust resource allocation to ensure that resources are effectively utilized during the peak period of user demand, avoiding resource waste or overload, thereby ensuring the quality of service while also improving the overall operating efficiency of the system. Secondly, the data request frequency and server load are analyzed in real time using deep learning algorithms, especially the dual-depth Q network algorithm, so that the data cache and distribution strategy can be intelligently adjusted according to the real-time changes in the network status, which not only improves the stability and reliability of data transmission, but also greatly reduces the delay caused by network congestion, ensuring that users can get a smoother access experience. In addition, through the optimized cache space priority table and targeted node layout construction, this solution can more accurately match the user's data access needs, whether in the accurate identification of user geographic location data or in the effective caching of expected access content data, it greatly improves the utilization of edge computing resources and the timeliness of content transmission.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

FIG1 is a schematic diagram of an embodiment of a content acceleration method based on edge caching in an embodiment of the present application;

FIG. 2 is a schematic diagram of an embodiment of a content acceleration device based on edge caching in an embodiment of the present application.

Detailed ways

Embodiments of the present application provide a content acceleration method, apparatus, device and storage medium based on edge caching. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

For ease of understanding, the specific process of the embodiment of the present application is described below. Please refer to Figure 1. An embodiment of the content acceleration method based on edge caching in the embodiment of the present application includes:

Step S101, collecting historical network traffic data, and extracting traffic characteristics from the historical network traffic data to obtain traffic characteristic data;

It is understandable that the execution subject of the present application may be a content acceleration device based on edge caching, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

Specifically, historical network traffic data includes but is not limited to website access logs, CDN (content distribution network) usage statistics, and outputs from network monitoring tools. For example, a video sharing platform wants to optimize the distribution of its video content to reduce buffering time and improve user satisfaction. Network traffic data from the past few months will be collected, including the number of video views, viewing time period, viewing duration, and geographic location of user requests. Taking the number of video views as an example, by analyzing the number of video views in different time periods, it is possible to identify which videos are particularly popular during specific time periods. For example, a newly released movie trailer may see a surge in views between 8 and 10 p.m. The traffic peak in this time period can be converted into a feature vector, indicating a high demand for new content among users.

In addition, viewing duration and the geographic location of user requests are also important aspects of extracting traffic features. For example, if it is found that users in a certain area tend to watch videos in full rather than simply skipping, this may indicate that users in that area have higher requirements for the quality of video content. After fusing these features, traffic feature data describing historical network traffic behavior is obtained. For example, based on traffic feature data, the demand for a certain type of video content in a certain period of time in the future can be predicted, and based on this, these contents can be cached preferentially on edge nodes close to the peak demand areas, thereby reducing the data transmission distance and time when users initiate requests, and accelerating the access speed of content.

Step S102: constructing an edge node resource allocation strategy based on traffic feature data, and generating a cache space priority table according to the edge node resource allocation strategy;

Specifically, based on the traffic characteristic data, it is possible to identify which content is "hot" content, that is, content that is in high demand and frequently accessed. Further analyze the demand patterns of these hot content, such as some content that may be particularly popular during specific time periods or in specific geographic areas. For example, for those TV series that are particularly popular on weekend evenings, their cache copies can be added in advance on edge nodes close to the locations of the main audience groups. Based on this, an edge node resource allocation strategy is formulated. The strategy takes into account the storage capacity, bandwidth capabilities of each edge node, and their relative relationship with the user's geographic location to ensure that hot content is prioritized and reasonably allocated to each edge node. Therefore, edge nodes with higher storage capacity and bandwidth capabilities and close to areas with high user density will be given higher cache priority.

According to the edge node resource allocation strategy, a cache space priority table is generated. The priority table clearly indicates which edge nodes each content item should be cached on and their cache priority on these nodes. For example, for popular TV series, the cache space priority table will indicate that they should be cached on all edge nodes close to the main audience groups, and give them a higher cache priority on these nodes. In this way, when users request to watch these TV series on weekend evenings, their requests can be responded to quickly because the relevant content has been cached on the edge nodes closest to them, which significantly reduces the delay of data transmission and improves the user experience. The edge node resource allocation strategy based on traffic feature data can not only ensure that hot content is effectively cached, but also optimize the resource utilization of the entire network and the access speed of users.

Step S103: collecting user data of the target user, and performing edge node matching on the user data through an edge node resource allocation strategy to obtain multiple initial edge nodes;

Specifically, data of the target user is collected, including the user's geographic location, requested video content ID, access time, etc. For example, a user in Beijing requests to watch a popular movie through a mobile client at 8 pm. In this process, the user's geographic location data can be obtained through the GPS function of the client, and the video content ID and access time are directly provided by the user's request.

Combine these collected user data with the existing edge node resource allocation strategy to match edge nodes. Based on historical network traffic data, it can reflect the user access patterns and content popularity changes in different geographical areas in different time periods. As described in the above example, edge nodes located in Beijing and its surrounding areas will be given priority at this time, and the suitability of each node as a cache node will be further evaluated based on the current server load of each node and the distance from the user's geographic location. If a node has a low current load and is close to the user, then this node will be given a higher matching score. Finally, an initial list of edge nodes is obtained, which are nodes that are considered to be the most suitable for processing the current request after a comprehensive evaluation based on the user's geographic location, the content characteristics of the request, and the current status of the node. For example, for the popular movie requested by the user in Beijing, an edge node located in the center of Beijing with a low current load and two other backup nodes located near the user's geographic location and also with a low load may be selected to form a multi-node cache and distribution strategy.

Step S104: constructing a node layout for multiple initial edge nodes based on the cache space priority table to obtain a target node layout;

Specifically, it should be noted that a video streaming service platform wants to improve the user's viewing experience through edge computing technology, especially to reduce video loading time. The platform has deployed multiple edge nodes around the world, which are distributed in areas with dense users, including major cities and population centers. Through the analysis of historical network traffic data, a detailed cache space priority table is obtained, which records the access frequency of different types of content in various time periods and geographical locations, as well as the performance indicators of each edge node. On this basis, the goal is to use this priority table to optimize the layout of existing edge nodes so that the most popular content can be cached on the nodes closest to the users. For example, if the cache space priority table shows that American and Korean dramas are particularly popular in Asia, while European users prefer to watch local TV shows and movies, then this information will directly affect the content caching strategy on Asian and European nodes.

Therefore, it is necessary to evaluate the current status of each initial edge node, including the node's storage capacity, bandwidth, and geographical distance from the user. Then, combined with the data in the cache space priority table, the platform can determine which content should be cached on which nodes first. In this process, a key decision is how to balance the load between nodes while ensuring that high-priority content can be cached on the node closest to the user.

Taking the Asian market as an example, considering the high popularity of American and Korean dramas, the platform may decide to increase the cache capacity of these contents on edge nodes close to user-dense areas. This may mean expanding the storage capacity of certain nodes or temporarily increasing bandwidth configuration during peak hours to cope with sudden traffic demands. At the same time, the platform also needs to consider the update frequency of content to ensure that newly released popular dramas can be distributed to key nodes in a timely manner. In addition, the construction of the target node layout also needs to take into account the content redundancy between nodes. In order to improve the robustness and reliability of the system, the same content may need to be backed up on multiple nodes. However, excessive redundancy will take up valuable storage resources, so it is necessary to optimize the redundancy strategy through intelligent algorithms to ensure the rationality of content distribution.

Step S105: Obtain a data access request from a target user, and perform access node matching on the data access request based on the target node layout to obtain a target access node.

Specifically, assume that an international news website wants to accelerate the global distribution of its content through edge caching technology. The website has multiple edge nodes distributed around the world, and each node caches part of the website content so that users can get data from the nearest node. When a user initiates an access request, for example, a user in region B tries to access an article about the latest international news, the user's approximate geographic location is first identified through the user's IP address. This information is then compared with the target node layout. In this example, the target node layout is based on the traffic feature data and edge node resource allocation strategy previously constructed, which details the geographic location of each node, the type and priority of the cached content, and the performance indicators of the node.

According to the target node layout, several edge nodes in Europe can be quickly identified. These nodes become the first choice for processing the request of the user in region B due to their geographical advantages. However, matching based on geographical location alone is not enough. The current load of the node and the cache status of the requested content also need to be considered. For example, if the nearest node is currently under high load, or the requested news article is not yet cached at the node, then another node with a better matching condition may need to be selected.

Furthermore, these potential nodes are comprehensively evaluated and matched through the Dual Deep Q Network (DDQN) algorithm. The DDQN algorithm calculates the expected efficiency of each node in processing the request by considering multiple factors such as the node load, the cache priority of the content, and the user's geographical location. In this way, the algorithm ultimately determines the best target access node. In our example, assume that an edge node located in area C, which is closer to area B, is selected as the target access node because the node currently has a moderate load and has cached the news article requested by the user, and can respond to the user's request at the fastest speed.

In the embodiment of the present application, the efficiency and quality of content transmission are significantly improved through resource allocation and data transmission strategies. First, by collecting and analyzing historical network traffic data, the constructed edge node resource allocation strategy can dynamically adjust resource allocation to ensure that resources are effectively utilized during the peak period of user demand, avoiding resource waste or overload, thereby ensuring the quality of service while also improving the overall operating efficiency of the system. Secondly, a deep learning algorithm, especially a dual-depth Q network algorithm, is used to analyze the data request frequency and server load in real time, so that the data cache and distribution strategy can be intelligently adjusted according to the real-time changes in the network status, which not only improves the stability and reliability of data transmission, but also greatly reduces the delay caused by network congestion, ensuring that users can get a smoother access experience. In addition, through the optimized cache space priority table and targeted node layout construction, this solution can more accurately match the user's data access needs, whether in the accurate identification of user geographic location data or in the effective caching of expected access content data, it greatly improves the utilization of edge computing resources and the timeliness of content transmission.

In a specific embodiment, the process of executing step S101 may specifically include the following steps:

(1) Collect historical network traffic data and traverse the historical network traffic data to obtain historical access frequency, historical access time, and historical access content type;

(2) Perform vector conversion on the historical access frequency to obtain a first vector, perform vector conversion on the historical access time to obtain a second vector, and at the same time, perform vector conversion on the historical access content type to obtain a third vector;

(3) performing vector fusion on the first vector and the second vector to obtain a first fused vector;

(4) performing vector fusion on the first vector and the third vector to obtain a second fused vector;

(5) performing vector fusion on the second vector and the third vector to obtain a third fused vector;

(6) Inputting the first fused vector into a preset convolutional long short-term memory network model to extract high-frequency access time features, thereby obtaining high-frequency access time features;

(7) Inputting the second fused vector into the convolutional long short-term memory network model to extract the high-frequency access content features, thereby obtaining the high-frequency access content features;

(8) Inputting the third fused vector into the convolutional long short-term memory network model to extract long-term access content and obtain long-term access content features;

(9) Fusing the high-frequency access time features and the high-frequency access content features to obtain a first fused feature, and fusing the high-frequency access content features and the long-term access content features to obtain a second fused feature;

(10) Inputting the first fused feature into the feature recognition layer of the convolutional long short-term memory network model to perform traffic pattern recognition to obtain the traffic pattern. Meanwhile, inputting the second fused feature into the feature recognition layer of the convolutional long short-term memory network model to perform content trend recognition to obtain the access content trend.

(11) Combine traffic patterns and access content trends into traffic feature data.

Specifically, suppose you are managing a large video streaming platform with edge nodes all over the world, with the goal of providing users with a low-latency, high-quality video viewing experience. To achieve this goal, you first need to collect the platform's historical network traffic data, which includes information such as the frequency of users accessing video content, the specific time of access, and the type of content accessed. For example, historical data may show that users from a specific geographic area access a popular series significantly more frequently on weekend evenings than at other time periods. After collecting these historical network traffic data, the task is to conduct an in-depth information traversal of these data and convert them into vector form for further analysis. Through in-depth analysis of historical access frequency, access time, and access content type, these data can be converted into three vectors: the first vector represents the historical access frequency, the second vector represents the historical access time, and the third vector represents the historical access content type. Furthermore, by fusing different combinations of the three basic vectors, three fusion vectors are obtained: the first fusion vector is a fusion of access frequency and access time, which can reveal the access pattern of specific content in different time periods; the second fusion vector is a fusion of access frequency and content type, which helps identify which types of content are more popular with users; and the third fusion vector is a fusion of access time and content type, indicating when different types of content receive more attention. In order to extract deep traffic features from these fusion vectors, they are input into a pre-set convolutional long short-term memory network (C-LSTM) model. This model combines the powerful expression ability of convolutional neural network (CNN) on image and sequence data and the processing ability of long short-term memory network (LSTM) on long-term dependency of time series data, making it particularly suitable for processing and analyzing time series data. Through the processing of the C-LSTM model, high-frequency access time features can be extracted from the first fusion vector, high-frequency access content features can be extracted from the second fusion vector, and long-term access content features can be extracted from the third fusion vector.

Finally, the high-frequency access time feature and the high-frequency access content feature are further fused to obtain the first fusion feature, which reflects which content is most popular at what time; at the same time, the high-frequency access content feature is fused with the long-term access content feature to obtain the second fusion feature, which reveals which content has sustained popularity. By inputting these fusion features into the C-LSTM model for analysis again, we finally get a comprehensive reflection of the traffic pattern and access content trend. These traffic feature data provide valuable guidance on how to optimize the content caching strategy of edge nodes.

For example, if it is found that science fiction and fantasy movies are particularly popular on the West Coast of the United States on weekend nights, and this trend has continued over the past few months, then such movies can be cached preferentially on edge nodes in the region to ensure that when users seek to watch these contents on weekend nights, they can obtain data from the geographically closest node, thereby significantly accelerating the loading speed of content and improving user experience. This method based on in-depth analysis of historical network traffic data and extraction of traffic characteristics enables smarter and more efficient management of edge cache resources, ultimately achieving the goal of accelerating content access.

It should be noted that the first fusion vector can reveal the popularity of specific content in different time periods by combining access frequency and time information. The second fusion vector can not only predict which types of content will be frequently accessed, but also identify potential popular trends by fusing access frequency and content type, so as to prepare for content caching in advance. The third fusion vector can identify the content type with increasing demand in a specific time period by combining the characteristics of access time and content type, thereby optimizing the caching strategy for specific time periods such as morning peak or evening peak, ensuring that users can quickly access the content of interest at any time.

In a specific embodiment, the process of executing step S102 may specifically include the following steps:

(1) Performing pattern recognition on the traffic pattern to obtain pattern recognition data, wherein the pattern recognition data includes: periodic traffic change value and traffic peak value;

(2) Use the time series prediction algorithm to predict the bandwidth demand of periodic traffic changes and traffic peaks to obtain the bandwidth demand for multiple time periods;

(3) Identify the peak periods of bandwidth demand in multiple time periods to obtain the peak traffic periods;

(4) Identify the low-peak periods of bandwidth demand in multiple time periods to obtain the low-peak periods of traffic;

(5) Identify high-demand content types based on access content trends and obtain a high-demand content type set;

(6) Generate an initial resource allocation strategy based on the peak traffic hours and the set of high-demand content types;

(7) Modify the initial resource allocation strategy during the low traffic period to obtain the edge node resource allocation strategy;

(8) Prioritize the high-demand content type set through the edge node resource allocation strategy to obtain sorted data, and generate a cache space priority table based on the sorted data.

Specifically, through the convolutional long short-term memory network (C-LSTM) model, the collected historical network traffic data is pattern-recognized, and two key information can be extracted from it: periodic traffic change value and traffic peak value. It reflects the regularity and irregular peaks of user access to content, and provides a basis for understanding traffic fluctuations. For example, suppose an online education platform experiences a traffic peak between 8 and 10 pm every Monday. This pattern may be due to the platform launching popular live courses during this time period. Using the long short-term memory network (LSTM), the bandwidth demand for a specific period of time in the future is predicted based on historical periodic traffic changes and traffic peaks. Not only based on past traffic patterns, but also considering possible trends and seasonal factors. In this way, it can be predicted that the traffic peak every Monday night will continue in the next few months, or even become more significant.

Peak traffic periods and trough traffic periods are identified from bandwidth demands in multiple time periods. By comparing the predicted bandwidth demands in different time periods, the periods with significantly higher than average demands are identified as peak traffic periods, while the periods with significantly lower than average demands are identified as trough traffic periods. At the same time, the analysis of access content trends helps identify content types with high user demand and form a set of high-demand content types. For example, a sudden increase in news reports due to recent hot events, or topic content that has become popular due to seasonal factors. Based on the above content, an initial resource allocation strategy is generated to guide how to prioritize the allocation of cache resources for edge nodes, taking into account peak traffic periods and high-demand content type sets. For example, if the forecast shows that technology news will have a peak in traffic every Monday night, then the relevant edge nodes will be designated to prioritize caching of such content.

However, in order to ensure the overall efficiency and robustness of the system, it is also necessary to consider the impact of low-traffic periods on the initial resource allocation strategy and make policy corrections accordingly. This correction aims to balance resource utilization in different time periods and ensure that precious cache resources are not wasted during low-traffic periods. Finally, through the edge node resource allocation strategy, the set of high-demand content types is prioritized and a cache space priority table is generated accordingly. This table directly guides which content should be cached first and on which edge nodes to ensure that users can access the content as quickly as possible.

In a specific embodiment, the process of executing step S103 may specifically include the following steps:

(1) Collecting user data of target users, where the user data includes user geographic location data and data on the content that the user is expected to access;

(2) Identify the server computing area based on the user's geographic location data to obtain the server computing area;

(3) Collect data on edge nodes in the server computing area to obtain regional node data;

(4) Fusing user geographic location data and user expected content data to obtain user demand fingerprints;

(5) Perform node resource traversal on regional node data to obtain node computing resources corresponding to multiple regional nodes. At the same time, collect resource load data of multiple regional nodes;

(6) Based on the user demand fingerprint, the resource load data of multiple regional nodes are screened to obtain multiple initial edge nodes.

Specifically, user data includes the user's geographic location data and the content data that the user intends to access. For example, when a user in region A tries to watch a newly released Hollywood movie, his geographic location is obtained as region A through the user's device. At the same time, the user searches or clicks on the link of the movie, so the content that the user intends to access is known.

The server computing area is identified based on the user's geographic location data. This step aims to determine the best service area for the user so that the most suitable edge node can be assigned to the user's request. In the above example, the user is identified as being in area A, and then area A and its surrounding areas are identified as server computing areas. Subsequently, detailed data collection is performed on all edge node information in the server computing area. This includes the geographic location of each node, available computing resources, and current resource load conditions, and ultimately a comprehensive regional node data view is obtained.

Then, the user's geographic location data and expected access content data are fused to obtain the user demand fingerprint. The user demand fingerprint reflects the user's specific needs and preferences, including the type of content the user wants to access and the geographic location from which they access the content. In the above example, the user demand fingerprint will contain information that the user is located in region A and that they intend to access new movies. After that, the resources of all edge nodes in the region are traversed, including the computing power resources of each node and the current resource load. Finally, based on the user demand fingerprint, the resource load data of multiple regional nodes is intelligently screened to determine the initial edge node that best suits the user's needs. This screening process takes into account the node's geographic location, available resources, and current load. In the above example, an edge node located in region A with a relatively low current load and that has cached the required movie content can be selected.

In a specific embodiment, the process of executing step S105 may specifically include the following steps:

(1) Obtain the target user's data access request and parse it to obtain the data request frequency and current server load data;

(2) Obtain the field of view data of the target user and collect the popularity value of the video content to be accessed. Input the field of view data and the popularity value set into the preset deep deterministic policy gradient algorithm to predict the focus area of the video content and obtain the user focus area.

(3) extracting the content to be cached from the video content to be accessed based on the user's focus area to obtain the content to be cached;

(4) Data caching is performed on the content to be cached through the target node layout. Based on the target node layout, the data request frequency and the current server load data are input into the preset dual deep Q network algorithm for access node matching to obtain the target access node.

Specifically, assuming that you are operating a popular video streaming service, the challenge is how to quickly respond to users' video content requests worldwide, especially during peak hours when there are a large number of users. First, you need to capture the data access requests of target users, including the frequency of user video requests and real-time monitoring of the current server load. For example, a user may frequently try to access a newly released movie, and the current load data shows that the server is facing high traffic pressure. By analyzing this information, you can understand the user's access intention and the current processing capacity of the server.

Obtaining the user's field of view data can help predict the part of the video that the user is most interested in. At the same time, the popularity value of the video content to be accessed is collected. The popularity value is based on the number of views, sharing frequency or user ratings of the video. After inputting the field of view data and popularity value into the preset deep deterministic policy gradient (DDPG) algorithm, the user's focus area is predicted, that is, the part of the video content that the user is most likely to pay attention to. For example, if a large number of users focus on the same exciting clip when watching a movie, this focus area can be identified.

Next, highly-watched segments are extracted from the entire video content as the content to be cached. Therefore, not only the entire movie needs to be cached, but those particularly popular segments should be prioritized to ensure that they can be loaded quickly when users access them.

Finally, based on the target node layout, the data request frequency and current server load data will be combined to intelligently match all available edge nodes using the preset dual deep Q network (DDQN) algorithm to determine the edge node that is most suitable for caching the content to be cached. Based on the node's geographic location, current load, and network latency between the node and the user, ensure that the selected node can both bear the new load and provide the fastest content access speed. For example, if a node in region D is identified to have a low current load and is closest to the user's geographic location requesting a movie, then this node will be selected as the target node for caching the focus segment of the movie.

In a specific embodiment, the process of inputting the field of view data and the popularity value set into a preset deep deterministic policy gradient algorithm to predict the video content focus area may specifically include the following steps:

(1) Obtain the field of view angle data of the target user, and extract the video views of the video content to be accessed to obtain the video views;

(2) Extract the video likes for the video content to be visited to obtain the video likes;

(3) Extract the complete viewing volume of the video content to be accessed to obtain the complete viewing volume of the video;

(4) Using a preset weight parameter set, the video popularity value is calculated for the video views, video likes, and video complete views to obtain a popularity value set;

(5) Calculating the video frame ratio value of the field of view angle data to obtain the video frame ratio value corresponding to the field of view angle data;

(6) Based on the video frame ratio value, the field of view angle data and the popularity value set are input into the deep deterministic policy gradient algorithm, and the coordinates of the center point of the focus area are calculated by the center point coordinate calculation formula to obtain the coordinates of the center point of the focus area. The center point coordinate calculation formula is as follows:

;

Where n is the number of data points of the field of view data, is the coordinate of the center point of the focal area, is the coordinate of the i-th field of view data point, is the angular offset of the i-th field of view data point relative to the center of the video content, is the first exponential parameter, is the second exponential parameter, is the weight factor, where The calculation formula is as follows;

;

in, is the popularity value corresponding to the i-th viewing angle data point, is the Euclidean distance from the i-th field of view data point to the center of the video content, is the correction factor;

(7) Generate multiple candidate focus areas based on the coordinates of the center point of the focus area, calculate the reward score for each candidate focus area using a preset reward function, and obtain reward score data for each candidate focus area;

(8) Based on the reward score data of each candidate focus area, multiple candidate focus areas are screened to obtain the user focus area.

Specifically, suppose you manage a video sharing platform and hope to accelerate the access of video content through edge caching technology. First, the platform needs to obtain the field of view data of the target user, which can be collected by analyzing the interactive information such as the mouse movement, scrolling behavior, or touch behavior on the mobile device when the user is watching the video. For example, when a user is watching a science fiction movie, it is possible to record the user's behavior of pausing or repeating the viewing at a specific scene, thereby capturing the user's field of view data. Perform a multi-dimensional popularity analysis on the video content to be accessed. First, through the video view extraction, the total number of times the video has been viewed can be known; then, the video like extraction reflects another dimension of the video's popularity; finally, the video complete view extraction reveals the completeness of the video content watched by the audience. Together, these indicators constitute a set of popularity values for the video content, providing comprehensive video popularity information. For example, a newly launched short film has a high number of views and likes in a short period of time, but a low number of complete views, which may indicate that users are highly curious about the short film, but the content may not continue to attract users to watch.

The field of view data and the video popularity value set are combined and analyzed using the deep deterministic policy gradient (DDPG) algorithm. The center point coordinate calculation formula accurately predicts the part of the video content that the user is most interested in, that is, the center point coordinate of the focus area. For example, if the analysis shows that the user concentrates a lot of field of view data when watching a specific explosion scene in a science fiction movie, and the popularity value of this scene is high, this scene can be identified as the user's focus area. Based on the center point coordinates of the focus area, multiple candidate focus areas are generated, and the reward scores of these candidate areas are calculated through the preset reward function. This step aims to evaluate the relevance and attractiveness of each candidate focus area to ensure that the focus area that best represents the user's interest is selected. For example, for a video containing multiple highly popular scenes, the reward score will be calculated for each scene, and the scene with the highest score will be selected as the focus area. Finally, the reward score data of each candidate focus area is screened to determine the final user focus area. Based on this focus area, the cached content is further extracted, and its cache location on the edge node is optimized according to the target node layout. For example, after analyzing and determining that several highly popular explosion scenes in science fiction movies are focus areas, these scenes are preferentially cached on the edge nodes closest to the target user's geographic location so that users can access these exciting contents with the lowest latency.

In a specific embodiment, the process of inputting the data request frequency and the current server load data into the preset dual-depth Q network algorithm to perform the access node matching step may specifically include the following steps:

(1) Analyze the network connection status of the target node layout to obtain the current network connection status;

(2) matching a data caching strategy based on the network connection status, and caching the cached content through the data caching strategy, wherein the data caching strategy includes a data allocation sub-strategy and a data transmission sub-strategy;

(3) Inputting the data request frequency into the dual deep Q network algorithm to construct the frequency state vector and obtain a multi-dimensional frequency state vector;

(4) Input the current server load data into the dual deep Q network algorithm to construct the load state vector and obtain a multi-dimensional load state vector;

(5) Input the multi-dimensional frequency state vector and the multi-dimensional load state vector into the main network of the dual deep Q network algorithm to execute the action, and obtain the action execution state data and reward parameters;

(6) Input the action execution state data and reward parameters into the target network of the dual deep Q network algorithm to perform maximum Q value dynamic analysis to obtain the maximum Q value action;

(7) Extract the target Q value of the action with the maximum Q value, match the access nodes of the target node layout based on the target Q value, and obtain the target access node.

Specifically, the current network connection status is obtained by analyzing the network connection status of the target node layout. This step is crucial because the fluctuation of the network status directly affects the efficiency and stability of content transmission. For example, in a typical application scenario, if a user tries to access a popular video in the edge cache, the network connection status with the target edge node is first checked, including but not limited to bandwidth usage, latency, and packet loss rate. These parameters can reflect the network congestion and transmission quality in real time, providing a basis for subsequent decisions. Based on the obtained network connection status, the data caching strategy is matched, including the data allocation sub-strategy and the data transmission sub-strategy. For example, when the network status is good, the data transmission sub-strategy may choose a higher transmission rate to reduce the user's waiting time. On the contrary, when the network condition is poor, a more conservative transmission strategy may be adopted, such as reducing the transmission rate, to avoid further deterioration of network congestion.

The frequency of data requests and the current server load data are used as inputs, and the frequency state vector and load state vector are constructed using the dual deep Q network algorithm. These two vectors represent the user's data request mode and the current workload of the edge node, respectively, and are important bases for optimization decisions. The dual deep Q network algorithm can predict the long-term benefits of taking different actions in a specific state by learning from past experience, so as to make the best choice. With the frequency state vector and the load state vector as inputs, the main network of the dual deep Q network algorithm will execute the action and output the action execution state data and reward parameters. This step is the key to finding the optimal cache and transmission strategy. The algorithm will try different strategy combinations and finally determine the strategy that best suits the current network and server status through continuous trial and error learning. Finally, the action execution state data and reward parameters will be sent to the algorithm's target network for dynamic analysis of the maximum Q value to determine the best action. This optimal action represents the cache and transmission strategy that can maximize long-term benefits under the current state. For example, for the request for the popular video mentioned earlier, the best action may include selecting an edge node that is closest to the user and has the best network connection status for content distribution, or adjusting the video bit rate to ensure smooth playback when the network status is poor.

The above describes the content acceleration method based on edge cache in the embodiment of the present application. The following describes the content acceleration device based on edge cache in the embodiment of the present application. Please refer to Figure 2. An embodiment of the content acceleration device based on edge cache in the embodiment of the present application includes:

The collection module 201 is used to collect historical network traffic data and extract traffic characteristics from the historical network traffic data to obtain traffic characteristic data;

A generating module 202, configured to construct an edge node resource allocation strategy based on the traffic characteristic data, and generate a cache space priority table according to the edge node resource allocation strategy;

A matching module 203 is used to collect user data of a target user, and perform edge node matching on the user data according to the edge node resource allocation strategy to obtain a plurality of initial edge nodes;

A construction module 204 is used to construct a node layout for the plurality of initial edge nodes based on the cache space priority table to obtain a target node layout;

The acquisition module 205 is used to acquire the data access request of the target user, and perform access node matching on the data access request based on the target node layout to obtain a target access node.

Through the collaboration of the above components, the efficiency and quality of content transmission are significantly improved through resource allocation and data transmission strategies. First, by collecting and analyzing historical network traffic data, the constructed edge node resource allocation strategy can dynamically adjust resource allocation to ensure that resources are effectively utilized during the peak period of user demand, avoiding resource waste or overload, thereby ensuring the quality of service while improving the overall operation efficiency of the system. Secondly, deep learning algorithms, especially the dual deep Q network algorithm, are used to analyze the data request frequency and server load in real time, so that the data cache and distribution strategy can be intelligently adjusted according to the real-time changes in the network status, which not only improves the stability and reliability of data transmission, but also greatly reduces the delay caused by network congestion, ensuring that users can get a smoother access experience. In addition, through the optimized cache space priority table and targeted node layout construction, this solution can more accurately match the user's data access needs, whether in the accurate identification of user geographic location data or the effective caching of expected access content data, which greatly improves the utilization of edge computing resources and the timeliness of content transmission.

The present application also provides a content acceleration device based on edge cache, and the content acceleration device based on edge cache includes a memory and a processor, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the content acceleration method based on edge cache in the above-mentioned embodiments.

The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the edge cache-based content acceleration method.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (10)

1. The content acceleration method based on the edge cache is characterized by comprising the following steps of:

collecting historical network flow data, and extracting flow characteristics of the historical network flow data to obtain flow characteristic data;

Constructing an edge node resource allocation strategy according to the flow characteristic data, and generating a cache space priority table according to the edge node resource allocation strategy;

collecting user data of a target user, and performing edge node matching on the user data through the edge node resource allocation strategy to obtain a plurality of initial edge nodes;

based on the buffer space priority table, carrying out node layout construction on a plurality of initial edge nodes to obtain a target node layout;

and acquiring the data access request of the target user, and performing access node matching on the data access request based on the target node layout to obtain a target access node.

2. The content acceleration method based on edge cache as set forth in claim 1, wherein the collecting historical network traffic data and extracting traffic features of the historical network traffic data to obtain traffic feature data includes:

collecting the historical network flow data, and performing information traversal on the historical network flow data to obtain historical access frequency, historical access time and historical access content types;

Performing vector conversion on the historical access frequency to obtain a first vector, performing vector conversion on the historical access time to obtain a second vector, and performing vector conversion on the historical access content type to obtain a third vector;

vector fusion is carried out on the first vector and the second vector to obtain a first fusion vector;

vector fusion is carried out on the first vector and the third vector to obtain a second fusion vector;

vector fusion is carried out on the second vector and the third vector, and a third fusion vector is obtained;

Inputting the first fusion vector into a preset convolution long-short-term memory network model to extract high-frequency access time characteristics, so as to obtain the high-frequency access time characteristics;

Inputting the second fusion vector into the convolution long-short-term memory network model to extract high-frequency access content characteristics, so as to obtain the high-frequency access content characteristics;

Inputting the third fusion vector into the convolution long-short-term memory network model to extract long-term access content, so as to obtain long-term access content characteristics;

performing feature fusion on the high-frequency access time feature and the high-frequency access content feature to obtain a first fusion feature, and performing feature fusion on the high-frequency access content feature and the long-time access content feature to obtain a second fusion feature;

inputting the first fusion features into a feature recognition layer of the convolution long-short-term memory network model to perform flow pattern recognition to obtain a flow pattern, and simultaneously inputting the second fusion features into a feature recognition layer of the convolution long-short-term memory network model to perform content trend recognition to obtain an access content trend;

And merging the flow mode and the access content trend into the flow characteristic data.

3. The content acceleration method based on edge cache as set forth in claim 2, wherein the constructing an edge node resource allocation policy from the traffic feature data and generating a cache space priority table according to the edge node resource allocation policy includes:

Performing pattern recognition on the flow pattern to obtain pattern recognition data, wherein the pattern recognition data comprises: periodic flow change values and flow peaks;

Predicting the bandwidth demand quantity of the periodic flow variation value and the flow peak value through a time sequence prediction algorithm to obtain the bandwidth demand quantity of a plurality of time periods;

carrying out peak time period identification on the bandwidth demand of a plurality of time periods to obtain a flow peak time period;

Carrying out low-valley period identification on the bandwidth demand of a plurality of periods to obtain a flow low-valley period;

identifying the high-demand content type of the access content trend to obtain a high-demand content type set;

Generating an initial resource allocation policy based on the traffic rush hour and the set of high demand content types;

Performing policy correction on the initial resource allocation policy through the traffic trough period to obtain the edge node resource allocation policy;

And carrying out priority ranking on the high-demand content type set through the edge node resource allocation strategy to obtain ranking data, and generating a buffer space priority table through the ranking data.

4. The content acceleration method based on edge cache as set forth in claim 1, wherein the collecting the user data of the target user and performing edge node matching on the user data by the edge node resource allocation policy to obtain a plurality of initial edge nodes includes:

Collecting user data of a target user, wherein the user data comprises user geographic position data and content data which the user expects to access;

Carrying out server calculation region identification on the user geographic position data to obtain a server calculation region;

acquiring data of edge node information in the server calculation area to obtain area node data;

Carrying out data fusion on the user geographic position data and the user expected access content data to obtain user demand fingerprints;

Traversing node resources of the regional node data to obtain node computing power resources corresponding to a plurality of regional nodes, and collecting resource load data of the plurality of regional nodes;

and carrying out node screening on the resource load data of a plurality of regional nodes through the user demand fingerprints to obtain a plurality of initial edge nodes.

5. The content acceleration method based on edge cache as set forth in claim 1, wherein the obtaining the data access request of the target user and performing access node matching on the data access request based on the target node layout to obtain a target access node includes:

acquiring a data access request of the target user, and analyzing the data access request to obtain data request frequency and current server load data;

acquiring the field angle data of the target user, collecting popularity values of video contents to be accessed, inputting the field angle data and the popularity value set into a preset depth deterministic strategy gradient algorithm to predict a video content focus area, and obtaining a user focus area;

Extracting the content to be cached from the video content to be accessed based on the user focus area to obtain the content to be cached;

And carrying out data caching on the content to be cached through the target node layout, and inputting the data request frequency and the current server load data into a preset dual-depth Q network algorithm to carry out access node matching based on the target node layout so as to obtain a target access node.

6. The content acceleration method based on edge cache according to claim 5, wherein the obtaining the view angle data of the target user, collecting popularity values of the video content to be accessed, inputting the view angle data and the popularity value set into a preset depth deterministic strategy gradient algorithm to predict a focus area of the video content, and obtaining a focus area of the user includes:

acquiring the field angle data of the target user, and extracting the video browsing amount of the video content to be accessed to obtain the video browsing amount;

Extracting the video praise amount of the video content to be accessed to obtain the video praise amount;

extracting the video complete viewing quantity of the video content to be accessed to obtain the video complete viewing quantity;

Carrying out video popularity value calculation on the video browsing amount, the video praise amount and the video complete watching amount through a preset weight parameter set to obtain a popularity value set;

Calculating a video frame proportion value of the field angle data to obtain a video frame proportion value corresponding to the field angle data;

Inputting the view angle data and the popularity value set into the depth certainty strategy gradient algorithm based on the video frame proportion value, and calculating the center point coordinates of the focus area through a center point coordinate calculation formula to obtain the center point coordinates of the focus area, wherein the center point coordinate calculation formula is as follows:

;

where n is the number of data points of the field angle data, Is the center point coordinates of the focal region,For the coordinates of the ith field angle data point,Is the angular offset of the ith field of view data point relative to the center of the video content,As a first index parameter of the first set of values,As a parameter of the second index of the light,Is a weight factor, wherein the weight factorThe calculation formula of (2) is shown below;

;

wherein, Is the popularity value corresponding to the ith field angle data point,Is the euclidean distance of the ith field angle data point to the center of the video content,Is a correction coefficient;

generating a plurality of candidate focus areas based on the focus area center point coordinates, and calculating the reward score of each candidate focus area through a preset reward function to obtain reward score data of each candidate focus area;

and carrying out region screening on a plurality of candidate focus regions based on the reward score data of each candidate focus region to obtain the user focus region.

7. The content acceleration method based on edge cache according to claim 5, wherein the performing data caching on the content to be cached by the target node layout, performing access node matching on the data request frequency and the current server load data input preset dual-depth Q network algorithm based on the target node layout, to obtain a target access node, includes:

analyzing the network connection state of the target node layout to obtain the current network connection state;

Based on the network connection state matching data caching strategy, carrying out data caching on the content to be cached through the data caching strategy, wherein the data caching strategy comprises a data distribution sub-strategy and a data transmission sub-strategy;

inputting the data request frequency into the dual depth Q network algorithm to construct a frequency state vector, so as to obtain a multidimensional frequency state vector;

inputting the current server load data into the dual-depth Q network algorithm to construct a load state vector, so as to obtain a multi-dimensional load state vector;

Inputting the multidimensional frequency state vector and the multidimensional load state vector into a main network of the dual-depth Q network algorithm for action execution to obtain action execution state data and rewarding parameters;

Inputting the action execution state data and the rewarding parameter into a target network of the dual-depth Q network algorithm to perform maximum Q value dynamic analysis, so as to obtain maximum Q value action;

And extracting a target Q value of the maximum Q value action, and performing access node matching on the target node layout based on the target Q value to obtain the target access node.

8. An edge cache-based content acceleration apparatus, comprising:

The acquisition module is used for acquiring historical network flow data, and extracting flow characteristics of the historical network flow data to obtain flow characteristic data;

The generating module is used for constructing an edge node resource allocation strategy according to the flow characteristic data and generating a buffer space priority table according to the edge node resource allocation strategy;

The matching module is used for collecting user data of a target user, and carrying out edge node matching on the user data through the edge node resource allocation strategy to obtain a plurality of initial edge nodes;

The construction module is used for constructing node layouts of a plurality of initial edge nodes based on the buffer space priority table to obtain target node layouts;

the acquisition module is used for acquiring the data access request of the target user, and carrying out access node matching on the data access request based on the target node layout to obtain a target access node.

9. An edge cache-based content acceleration apparatus, characterized in that the edge cache-based content acceleration apparatus comprises: a memory and at least one processor, the memory having instructions stored therein;

The at least one processor invoking the instructions in the memory to cause the edge cache based content acceleration device to perform the edge cache based content acceleration method of any of claims 1-7.

10. A computer readable storage medium having instructions stored thereon, which when executed by a processor implement the edge cache based content acceleration method of any of claims 1-7.