PCDN service identification method, local node, CDN server, equipment and storage medium

Abstract

The invention provides a PCDN service identification method, a local node, a CDN server, electronic equipment and a computer readable storage medium, and relates to the technical field of network detection. The identification method comprises the following steps: establishing a PCDN service identification global model through combination of a CDN server and other local nodes in a plurality of local nodes based on federated learning; superposing PCDN service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model; and identifying own network flow based on the PCDN service identification personalized model so as to identify a PCDN service user. The problem that PCDN service identification accuracy is not high in the prior art is at least solved. The method is suitable for network detection and service detection scenes.

Description

Technical Field

[0001] The present invention relates to the field of network detection technology, and in particular to a PCDN service identification method, a local node, a CDN server, an electronic device and a computer-readable storage medium. Background Art

[0002] Since PCDN (Peer to Peer Content Delivery Network, also known as CDN based on P2P technology) technology directly places content on the user's mobile terminal or router, the mobile terminal or router becomes a larger "content source", which brings huge risks and losses to the rights and interests of operators and users, as follows:

[0003] (1) The terminal becomes a PCDN node and interacts with other devices, which may lead to the leakage of private data or security vulnerabilities and malicious attacks.

[0004] (2) Massive PCDN traffic creates tremendous pressure on operators’ backbone transmission networks, affecting the overall user perception and experience.

[0005] The traditional method used by operators to identify PCDN services is to use DPI (deep

[0006] PCDN services are identified by using deep packet inspection (DPI) traffic identification and traffic anomaly detection methods. DPI traffic identification and traffic anomaly detection methods include rule identification and AI identification. The rule identification algorithm usually combines the historical behavior of broadband users and uses specific defined rules for identification. The AI ​​identification method is to model the historical behavior data of broadband users, and then judge the real-time behavior of users through the model.

[0007] However, operators' traditional PCDN service identification methods require centralized data collection, analysis and processing, which requires high computing power, but the accuracy of PCDN service identification is not high. It also requires frequent data transmission and interaction, which poses a data security risk. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a PCDN service identification method, a local node, a CDN server, an electronic device and a computer-readable storage medium in view of the above-mentioned deficiencies in the prior art. The method can realize efficient and accurate PCDN service identification, improve the data analysis capability of the PCDN service identification model, improve the accuracy and efficiency of the analysis and processing of common feature data and personalized feature data, reduce maintenance and update costs, and realize more accurate and personalized PCDN service identification services.

[0009] In a first aspect, the present invention provides a method for identifying a point-to-point content distribution network (PCDN) service, which is applied to any local node among multiple local nodes of a target area content distribution network (CDN). The method for identifying a PCDN service includes: establishing a PCDN service identification global model through a CDN server and other local nodes among the multiple local nodes based on federated learning; training the PCDN service identification global model by superimposing PCDN service identification personalized feature data to obtain a PCDN service identification personalized model; and identifying its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0010] Preferably, before the PCDN service identification global model is jointly established through the CDN server and other local nodes in the multiple local nodes based on federated learning, the PCDN service identification method also includes: encrypting the local service data collected by itself and sending it to the CDN server; receiving PCDN service identification feature data sent by the CDN server, wherein the PCDN service identification feature data is obtained by the CDN server aligning the encrypted local service data with the BSS service data encrypted by the BSS node based on user ID matching, and the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data; and using the PCDN service identification common feature data as training data for establishing the PCDN service identification global model.

[0011] Preferably, the method of establishing a PCDN service identification global model through the CDN server and other local nodes in the multiple local nodes based on federated learning specifically includes: training the PCDN service identification model based on the PCDN service identification common feature data to obtain a local PCDN service identification model; sending the model parameters of the trained local PCDN service identification model to the CDN server, so that the CDN server aggregates the model parameters of the multiple local nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, where N is a positive integer; and receiving the PCDN service identification global model sent by the CDN server.

[0012] Preferably, the sending of the model parameters of the trained local PCDN service identification model to the CDN server so that the CDN server aggregates the model parameters of multiple nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, specifically includes: sending the model parameters of the local PCDN service identification model after the i-th round of training to the CDN server so that the CDN server performs weighted summation on the model parameters of multiple local nodes received in the i-th round to generate the optimized model parameters of the i-th round, wherein i=1, 2, ..., N, where N represents the round threshold; receiving The CDN server sends the optimized model parameters of the i-th round, adjusts the local PCDN service identification model after the i-th round of training, and obtains the local PCDN service identification model after the i+1-th round of training; determines whether the local PCDN service identification model after the i+1-th round of training meets the convergence state; in response to the local PCDN service identification model after the i+1-th round of training meeting the convergence state, sends the model parameters of the local PCDN service identification model after the i+1-th round of training to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes to generate a PCDN service identification global model.

[0013] Preferably, the BSS service data includes a user ID and at least one of the following: user real-name registration information, user location data, package type, consumption amount information, and place of origin information. Before the local service data collected by itself is encrypted and sent to the CDN server, the PCDN service identification method also includes: collecting local service data, wherein the local service data includes a user ID and at least one of the following: user service data, user signaling data, user service frequency, average daily upstream and downstream traffic of the service, service flow direction, and routing information.

[0014] In a second aspect, the present invention also provides a method for identifying point-to-point content distribution network PCDN services, which is applied to a CDN server of a target area content distribution network CDN. The method for identifying PCDN services includes: jointly establishing a PCDN service identification global model with multiple local nodes based on federated learning; sending the PCDN service identification global model to each local node, so that each local node superimposes PCDN service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model, and identifying its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0015] Preferably, before the PCDN service identification global model is jointly established with multiple local nodes based on federated learning, the PCDN service identification method also includes: receiving encrypted local service data sent by multiple local nodes, wherein the local service data is used to characterize the service data collected by the local node; obtaining BSS service data encrypted by the BSS node; aligning the encrypted local service data with the encrypted BSS service data based on user ID matching to obtain PCDN service identification feature data, wherein the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data; sending the PCDN service identification feature data to each local node, so that each local node uses the PCDN service identification common feature data as training data for training the local PCDN service identification model when establishing the PCDN service identification global model.

[0016] In a third aspect, the present invention also provides a local node, which is applied to a target area CDN, and includes a first establishment module, a training module and an identification module. The first establishment module is used to jointly establish a PCDN service identification global model with other local nodes in a plurality of local nodes through a CDN server based on federated learning. The training module is connected to the first establishment module and is used to superimpose service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model. The identification module is connected to the training module and is used to identify its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0017] In a fourth aspect, the present invention further provides a CDN server, which is applied to a target area CDN, and includes a second establishment module and a second sending module. The second establishment module is used to jointly establish a PCDN service identification global model with multiple local nodes based on federated learning. The second sending module is connected to the second establishment module and is used to send the PCDN service identification global model to each local node, so that each local node superimposes PCDN service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model, and identifies its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0018] In a fifth aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the PCDN service identification method provided in the first aspect or the second aspect.

[0019] In a sixth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for identifying the PCDN service provided in the first aspect or the second aspect above is implemented.

[0020] The present invention provides a method for identifying PCDN services, a local node, a CDN server, an electronic device, and a computer-readable storage medium. The CDN server combines the common feature data of PCDN services of multiple local nodes of the target area CDN to quickly establish a PCDN service identification global model with strong generalization ability, so as to distribute it to multiple local nodes of the target area CDN; and then trains the PCDN service identification global model through the personalized feature data of PCDN services of each local node of the target area CDN, and generates a PCDN service identification personalized model that meets the network traffic identification requirements of the local node itself, wherein the PCDN service identification personalized model can not only quickly analyze and process the common feature data of PCDN service identification, but also quickly analyze and process the personalized feature data of PCDN service identification, and then can quickly and accurately identify the PCDN service users of the local node itself. Therefore, the present invention can realize efficient and accurate PCDN service identification, improve the data analysis capability of the PCDN service identification model, improve the accuracy and efficiency of the analysis and processing of the common feature data and the personalized feature data, reduce the maintenance and update costs, and realize more accurate and personalized PCDN service identification services. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of a method for identifying a PCDN service according to Embodiment 1 of the present invention;

[0022] Figure 2 This is an example diagram of matching and aligning local service data with BSS service data in Embodiment 1 of the present invention;

[0023] Figure 3 This is a flow chart of a method for identifying a PCDN service according to Embodiment 2 of the present invention;

[0024] Figure 4 This is a flow chart of a method for identifying a PCDN service according to Embodiment 3 of the present invention;

[0025] Figure 5 This is a schematic diagram of the structure of a PCDN service identification device according to Embodiment 4 of the present invention;

[0026] Figure 6 This is a structural diagram of a PCDN service identification device according to Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0028] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0029] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0030] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0031] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0032] It can be understood that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0033] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0034] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0035] Embodiment 1:

[0036] like Figure 1 As shown, this embodiment provides a method for identifying a point-to-point content delivery network PCDN service, which is applied to any local node among multiple local nodes of a target area content delivery network CDN.

[0037] In this embodiment, the content delivery network CDN (Content Delivery Network) includes multiple areas, and there are several clients, i.e., local nodes, in each area. CDN is a regular edge computing network of the operator. It is a layer of intelligent virtual network built on the basis of the existing Internet. It mainly deploys a CDN server in each CDN area as the central server of its own area. The CDN server is responsible for data distribution and collection of local nodes in its own area. The CDN server collects and analyzes the DPI data and traffic monitoring data of all local nodes in its own area, generates the source station content of the CDN server, and distributes the source station content of the CDN server to all local nodes in its own area, so that users of local nodes can obtain the required content nearby. The source station content includes but is not limited to: analysis result information and storage information of DPI data and traffic monitoring data.

[0038] It should be noted that with the continuous development of optical communication technology, the broadband business of operators is not only a prerequisite for the development of the digital economy, but also one of the key engines driving the rapid growth of the digital economy. The broadband business of operators has strongly supported the extension and integration of the digital economy from the consumer Internet to the industrial Internet, and promoted the high-quality development of the economy and society. It is specifically reflected in the following aspects: ① Operators have achieved high coverage and high-speed access across the country through continuous investment and upgrading of broadband network infrastructure, providing solid physical layer support for the digital economy. ② The rapid development of broadband services has significantly improved network speed and stability, reduced unit bandwidth charges, and enabled enterprises and individual users to enjoy better network services at a lower cost, thereby promoting the prosperity of digital economic formats such as online business, remote office, online education, and video streaming. ③ High-speed broadband networks can support the application of new generation information technologies such as big data, cloud computing, the Internet of Things, 5G (5th Generation Mobile Communication Technology), and artificial intelligence, prompting traditional industries to achieve digital transformation and form new economic growth points. ④The core of the digital economy lies in data flow and processing. The broadband services provided by operators ensure the efficient transmission and storage of data, which is conducive to enterprises to explore data value and innovate business models. At the same time, it can also meet consumers' demand for real-time, high-definition and interactive digital services. ⑤By expanding the coverage of urban and rural broadband and lowering the access threshold, operators can help eliminate information asymmetry and digital divide between regions and share the economic and social benefits brought by digitalization. ⑥High-quality broadband services also strongly support the digitalization process in the fields of e-government, smart cities, telemedicine, and emergency response public services, and improve the effectiveness of social governance and the level of public services.

[0039] However, PCDN services also bring huge risks and losses to operators and users in terms of broadband business revenue. Therefore, it is urgent for operators to conduct strict inspections on PCDN services.

[0040] The PCDN service identification method includes:

[0041] S101, based on federated learning, a CDN server and other local nodes in a plurality of local nodes jointly establish a PCDN service identification global model.

[0042] In this embodiment, through a centralized federated learning framework, the local business data and centrally collected data of each local node in the target area CDN network are retained in the target area CDN network, and a virtual shared model is established by combining data from multiple participants to implement the PCDN service identification algorithm.

[0043] It should be noted that the federated learning algorithm includes but is not limited to: Federated Averaging (FedAvg), Federated stochastic gradient descent (FedSGD), Federated Proximal Algorithm (FedProx), Federated Recurrent Neural Network (Federated RNN), Hierarchical Federated Learning, Secure Multi-Party Computation (SMPC), Differential Privacy and Federated Transfer Learning. This embodiment takes the federated learning algorithm as the federated averaging algorithm as an example.

[0044] It should be noted that the federated average algorithm aggregates the model by training the model locally on each device and sending the model weights to the central server, and then performing an average update. Federated stochastic gradient descent uses only a single device for gradient updates in each round and sends the update results to the central server. Federated proximal optimization solves the problem of uneven data distribution among devices by adding a regularization term on the basis of FedAvg. Federated affine model refers to a federated learning model for processing sequence data, usually involving recurrent neural networks (RNN). Hierarchical federated learning is applicable to multi-level network architectures, such as federated learning between devices and edge servers. Secure multi-party computing ensures data privacy through cryptographic methods, allowing all parties to jointly calculate the model without exposing the data. Differential privacy adds noise during model training to protect the privacy of individual data. Federated transfer learning combines the idea of ​​transfer learning and allows model training on different data distributions.

[0045] Optionally, before S101: jointly establishing a PCDN service identification global model through the CDN server and other local nodes in the multiple local nodes based on federated learning, the PCDN service identification method further includes steps S104 to S106:

[0046] S104, encrypting the local business data collected by itself and sending it to the CDN server.

[0047] Optionally, before S104: encrypting the local service data collected by itself and sending it to the CDN server, the PCDN service identification method further includes:

[0048] S107, collecting local service data, wherein the local service data includes a user ID and at least one of the following: user service data, user signaling data, user service frequency, daily average uplink and downlink traffic of services, service flow direction, and routing information.

[0049] In this embodiment, the data involved in federated learning modeling includes local business data collected by each local node. When each local node accumulates its own local business data, it will not only identify the user ID of the business data, but also form a set of local business data distinguished by the user phone number by taking the user ID as the user phone number in this embodiment. When a local node collects its own local business data, it will also mark its own data source identifier for its own local business data. The data source identifier is used to characterize the source node of the local business data.

[0050] The local service data collected by each local node in this embodiment is generally user broadband service data of each category (enterprise, business district, residence), that is, user O domain information. The user O domain information includes but is not limited to: user service data, user signaling data, user service frequency, average daily upstream and downstream traffic of the service, service flow direction, and routing information.

[0051] Specifically, the BSS service data includes the user ID and at least one of the following: user real-name registration information, user location data, package type, consumption amount information, and location information.

[0052] In this embodiment, the data involved in the federated learning modeling also includes the BSS service data collected by the operator's BSS (Business Support System) node. Similarly, the BSS service data collected by the operator's BSS node also forms a BSS service data set distinguished by user telephone number.

[0053] The BSS service data collected by the BSS node in this embodiment is generally user B domain information, including but not limited to: user real-name registration information, user location data, package type, consumption amount information, and location information.

[0054] S105, receiving PCDN service identification feature data sent by the CDN server, wherein the PCDN service identification feature data is obtained by the CDN server aligning the encrypted local service data with the BSS service data encrypted by the BSS node based on user ID matching, and the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data.

[0055] In this embodiment, after collecting the local service data of each local node and the BSS service data of the BSS node, in order to avoid leaking data privacy, each local node and the BSS node encrypt their own local service data and BSS service data respectively. In addition, the prior art usually collects the local service data of each local node in a centralized manner, and identifies the service data of the PCDN service users of each local node through DPI traffic identification and traffic anomaly detection methods, such as rule identification and AI identification methods. Therefore, in this embodiment, the DPI traffic identification and traffic anomaly detection methods can also be preset on the CDN server. After each local node sends its own local service data to the CDN server respectively, and the BSS node sends the BSS service data to the CDN server, the CDN server can preliminarily identify the local service data of each local node and the BSS service data of the BSS node, the corresponding service data of the PCDN service users of each local node and the BSS node, and identify the corresponding service data of the PCDN service users of each local node and the BSS node through the preset DPI traffic identification and traffic anomaly detection method.

[0056] Then, the CDN server can confirm the common users of both parties without exposing their respective business data on the BSS node and all local nodes, and the business data of users that are not exposed or overlapped with each other, so as to combine the business data of the common users and realize the mutual supplementation of local business data and BSS business data under the common users of all local nodes and BSS nodes, so as to obtain the PCDN business identification feature data of all local nodes. Specifically, the CDN server adopts the Paillier homomorphic encryption algorithm to align and merge the encrypted local business data and BSS business data under the same user ID of all local nodes and BSS nodes, so as to obtain the PCDN business identification feature data of all local nodes, such as Figure 2 As shown. This embodiment matches and aligns local business data and BSS business data through the CDN server to obtain PCDN business identification feature data for the subsequent establishment of a PCDN business identification global model and a PCDN business identification personalized model, thereby realizing the fusion modeling of multi-domain data and the discrimination of the full amount of business data, and improving the accuracy of PCDN business user group identification. The Paillier homomorphic encryption algorithm is used to achieve matching and alignment of business data and BSS business data. The Paillier homomorphic encryption algorithm allows calculations to be performed directly on encrypted data without decryption, thereby protecting the privacy of sensitive business data during transmission and processing, reducing the risk of data leakage, and does not require frequent decryption and re-encryption processes, thereby reducing computational complexity, improving the response speed of the system, and being able to effectively support security and privacy requirements while maintaining data processing efficiency and accuracy.

[0057] Based on the data source identifier of the local business data in the PCDN business identification feature data, the PCDN business identification feature data of all local nodes are divided into the PCDN business identification feature data of each local node, and the PCDN business identification feature data of each local node are distributed to each local node. For different local nodes in the same regional CDN, the characteristics of the PCDN business identification feature data are that there is a lot of feature overlap and little user overlap. Therefore, before dividing the PCDN business identification feature data of all local nodes into the PCDN business identification feature data of each local node, this embodiment also aggregates the PCDN business identification feature data of all local nodes under the same regional CDN, determines the PCDN business identification common feature data and PCDN business identification personalized feature data of all local nodes under the same regional CDN, and uses the PCDN business identification common feature data of all local nodes under the same regional CDN to train the PCDN business identification global model, which can maximize the number of samples and improve the accuracy of the global model.

[0058] It should be noted that the common feature data for PCDN service identification refers to the feature data that is of common concern or common to all local nodes under the target area CDN, such as: traffic pattern data, user terminal feature data, geographic location distribution data, content type data and access frequency data. The personalized feature data for PCDN service identification refers to the unique user demand and behavior feature data of each local node under the target area CDN, such as: time sensitivity feature data, content caching strategy data, user portrait data within the node, user interaction behavior data and localized content preference data.

[0059] In addition to using the Paillier homomorphic encryption algorithm to achieve matching and alignment of business data and BSS business data, this embodiment can also achieve matching and alignment of business data and BSS business data through semi-homomorphic encryption (Partial Homomorphic Encryption, PHE), hierarchical homomorphic encryption (Liveled HE, LHE), and fully homomorphic encryption (Fully Homomorphic Encryption, FHE).

[0060] S106: Use the PCDN service identification common feature data as training data for establishing a PCDN service identification global model.

[0061] In this embodiment, after all local nodes obtain their own common feature data for PCDN service identification, the common feature data for PCDN service identification of all local nodes under the same regional CDN is used to establish and train the global model for PCDN service identification under the current regional CDN, so as to obtain a global model for PCDN service identification of the current regional CDN with strong generalization ability. In the case where there are many local nodes under the current regional CDN and the network traffic of the local nodes is large, the overall trend and characteristics of the feature data for PCDN service identification of all local nodes under the current regional CDN can be captured, and a global model for PCDN service identification can be quickly built. In this embodiment, by using the common feature data for PCDN service identification, a global model for PCDN service identification is built, and the model is distributed to multiple local nodes under the target regional CDN, and the generalization ability of the global model for PCDN service identification is used to quickly build a model for identifying PCDN service users at multiple local nodes under the target regional CDN.

[0062] Specifically, S101: establishing a PCDN service identification global model through the CDN server and other local nodes in the multiple local nodes based on federated learning, including steps S1011 to S1013:

[0063] S1011, training a PCDN service identification model based on PCDN service identification common feature data to obtain a local PCDN service identification model.

[0064] In this embodiment, the PCDN service identification model refers to a global model initially constructed by the CDN server based on a classification algorithm. The CDN server distributes the initially constructed global model to all local nodes of the current regional CDN, so that each local node trains the initially constructed global model separately to obtain a local PCDN service identification model.

[0065] It should be noted that the classification algorithms include but are not limited to: random forest, neural network, and support vector machine.

[0066] S1012, sending the model parameters of the trained local PCDN service identification model to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, where N is a positive integer.

[0067] In this embodiment, considering the differences between BSS service data and local service data, as well as the different local service data of local nodes under each regional CDN, in the federated learning architecture, each regional CDN adopts the same training model structure. Obviously, the model accuracy will be affected. Therefore, in this embodiment, a global PCDN service identification model that is adapted to all local nodes under each regional CDN is not constructed from the BSS node globally, but all local nodes under the same regional CDN perform training on the global PCDN service identification model. In this embodiment, the global PCDN service identification model of the current regional CDN is trained by all local nodes under the same regional CDN, and is transmitted to all local nodes of the current regional CDN for subsequent personalized training, which effectively improves the performance quality of the global model, and thus improves the accuracy of the global PCDN service identification model.

[0068] In addition, since the PCDN service identification common feature data of all local nodes under each regional CDN has different data volume scales and different overall distributions of data features, the contribution of the PCDN service identification common features of different local nodes to the global model is also different. Therefore, this embodiment trains the local PCDN service identification model by using the PCDN service identification common feature data at the local node, and then aggregates the model parameters of multiple local nodes based on federated learning through the CDN server to generate a PCDN service identification global model that meets the convergence state. By training the model at the local node, user data privacy can be effectively protected. Parallel training at multiple local nodes can significantly improve the speed of model training and reduce the computational burden of the CDN server. By aggregating model parameters instead of directly transmitting data, network bandwidth consumption can be effectively reduced, and the overall performance of the system can be improved. By continuously updating and optimizing the aggregated model, the accuracy and robustness of PCDN service identification can be improved.

[0069] Specifically, S1012: sending the model parameters of the trained local PCDN service identification model to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes received in N rounds based on federated learning, and generates a PCDN service identification global model that meets the convergence state, including: sending the model parameters of the local PCDN service identification model after the i-th round of training to the CDN server, so that the CDN server performs weighted summation on the model parameters of multiple local nodes received in the i-th round, and generates the optimized model parameters of the i-th round, where i=1, 2, ..., N, and N represents the round threshold. ; receiving the optimized model parameters of the i-th round sent by the CDN server, adjusting the local PCDN service identification model after the i-th round of training, and obtaining the local PCDN service identification model after the i+1-th round of training; judging whether the local PCDN service identification model after the i+1-th round of training meets the convergence state; in response to the local PCDN service identification model after the i+1-th round of training meeting the convergence state, sending the model parameters of the local PCDN service identification model after the i+1-th round of training to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes to generate a PCDN service identification global model.

[0070] In this embodiment, while the local node uses the common feature data of PCDN service identification to train the local PCDN service identification model, the local node will calculate the model parameters of its own local PCDN service identification model and send the model parameters to the CDN server of the CDN in the area where the local node is located. The model parameters include but are not limited to: gradient information, statistical information, encryption information, metadata, and convergence speed. In this embodiment, the model parameters are used as the gradient information of the loss function. For example, the gradient information of the loss function Used to characterize the adjustment requirement update of the model parameter ω when minimizing the loss function.

[0071] The CDN server of the CDN in the area where the local node is located receives the model parameters of all local nodes of the CDN in the area where the local node is located to construct a global model for PCDN service identification of the CDN in the area where the local node is located, specifically: ① weighted sum of the model parameters of all local nodes of the CDN in the area where the local node is located, and calculate the optimized model parameters of all local nodes of the CDN in all areas of the local node; ② distribute the optimized model parameters of all local nodes to all local nodes of the CDN in the area where the local node is located to update the local PCDN identification model of all local nodes; ③ for the model parameters obtained in different rounds, repeat ① and ② until the local PCDN identification model of all local nodes meets the convergence state, or reaches the threshold of the number of repeated executions, aggregate the model parameters of all local nodes of the CDN in the area where the local node is located, and generate a global model for PCDN service identification. In this embodiment, during the entire PCDN service identification global model construction process, the local node only communicates with the CDN server. In addition to the jointly maintained optimized model parameters, the local node cannot obtain any information of other local nodes of the CDN in its own area, thereby ensuring the confidentiality of the privacy data of multiple local nodes.

[0072] Calculate the optimization model parameters of all local nodes of all CDNs in the local node region, as shown in formula (1):

[0073]

[0074] Where N represents the total number of local nodes, n j represents the sample weight of the jth local node, n i Represents the sample weight of the i-th local node.

[0075] Aggregate the model parameters of all local nodes of the CDN in the area where the local node is located to generate a global model for PCDN service identification, as shown in formula (2):

[0076]

[0077] Where F(θ) service represents the global model of PCDN service identification of the CDN in the region where the local node is located. The CDN in the region where the local node is located includes K local nodes. k (θ) represents the local PCDN service identification model of the kth local node, n k represents the sample weight of the kth local node, and n represents the sum of the sample weights of the K local nodes. Therefore, the model parameters of the PCDN service identification global model are obtained by weighted summing the model parameters of each local PCDN service identification model according to the sample weight values.

[0078] It should be noted that the gradient information changes during the convergence process are as follows: In the early stages of training, the model loss may be high, the gradient information is generally large, pointing to a larger direction for adjusting the parameters, the model parameters are updated quickly, and the loss value drops significantly. As training progresses, the size of the gradient may gradually decrease, indicating that the loss is decreasing, and the model converges more slowly, and the pace of parameter updates may need to be adjusted to avoid over-updates. When the model is close to convergence, the gradient may be very small, indicating that the model has reached or is close to a local optimal point, and convergence may become very slow.

[0079] S1013: Receive a PCDN service identification global model sent by the CDN server.

[0080] S102, superimposing PCDN service identification personalized feature data to train a PCDN service identification global model to obtain a PCDN service identification personalized model.

[0081] In this embodiment, each local node will superimpose its own PCDN service identification common feature data based on its own PCDN service identification personalized feature data to train the received PCDN service identification global model, that is, on the basis of the training data of the PCDN service identification global model of each local node including its own PCDN service identification common feature data, its own PCDN service identification personalized feature data is also connected to the training data of the PCDN service identification global model for training the PCDN service identification global model, and finally a PCDN service identification personalized model adapted to localization is obtained. In this embodiment, a PCDN service identification personalized model adapted to localization is finally obtained by training the received PCDN service identification global model based on the PCDN service identification personalized feature data of the local node, thereby improving the accuracy of PCDN service identification.

[0082] S103: Identify the own network traffic based on the PCDN service identification personalized model to identify the PCDN service user.

[0083] The present embodiment provides a method for identifying PCDN services. A CDN server combines common feature data for identifying PCDN services of multiple local nodes of a target area CDN to quickly establish a global model for identifying PCDN services with strong generalization capability, and distributes the model to multiple local nodes of the target area CDN. The global model for identifying PCDN services is then trained using the personalized feature data for identifying PCDN services of each local node of the target area CDN, and a personalized model for identifying PCDN services that meets the network traffic identification requirements of the local node itself is generated. The personalized model for identifying PCDN services can not only quickly analyze and process common feature data for identifying PCDN services, but can also quickly analyze and process personalized feature data for identifying PCDN services, and can then quickly and accurately identify its own PCDN service users, thereby achieving efficient and accurate PCDN service identification, improving the data analysis capability of the PCDN service identification model, improving the accuracy and efficiency of analyzing and processing common feature data and personalized feature data, and reducing maintenance and update costs while achieving more accurate and personalized PCDN service identification services. In addition, by matching and aligning local service data and BSS service data through CDN servers, PCDN service identification feature data is obtained for the subsequent establishment of PCDN service identification global model and PCDN service identification personalized model, realizing the fusion modeling of multi-domain data and the discrimination of full service data, and improving the accuracy of PCDN service user group identification. The Paillier homomorphic encryption algorithm is adopted to achieve the matching and alignment of service data and BSS service data. The Paillier homomorphic encryption algorithm allows calculations to be performed directly on encrypted data without decryption, thereby protecting the privacy of sensitive service data during transmission and processing, reducing the risk of data leakage, and does not require frequent decryption and re-encryption processes, thereby reducing the complexity of calculations, improving the response speed of the system, and being able to effectively support security and privacy requirements while maintaining the efficiency and accuracy of data processing. By aggregating the PCDN service identification feature data of all local nodes under the same regional CDN, determining the common feature data and personalized feature data of PCDN service identification of all local nodes under the same regional CDN, and using the common feature data of PCDN service identification of all local nodes under the same regional CDN to train the PCDN service identification global model, the number of samples can be maximized and the accuracy of the global model can be improved. By using the common feature data of PCDN service identification to model the PCDN service identification global model and distributing it to multiple local nodes under the target regional CDN, and using the generalization ability of the PCDN service identification global model, the model for identifying PCDN service users in multiple local nodes under the target regional CDN can be quickly established.The PCDN service identification global model of the current regional CDN is trained through all local nodes under the same regional CDN, and is transmitted to all local nodes of the current regional CDN for subsequent personalized training, which effectively improves the performance quality of the global model, and then improves the accuracy of the PCDN service identification global model. By using the common feature data of PCDN service identification in the local node to train the local PCDN service identification model, and then aggregating the model parameters of multiple local nodes based on federated learning through the CDN server, a PCDN service identification global model that meets the convergence state is generated. By training the model in the local node, user data privacy can be effectively protected. Parallel training on multiple local nodes can significantly improve the speed of model training and reduce the computational burden of the CDN server. By aggregating model parameters instead of directly transmitting data, network bandwidth consumption can be effectively reduced and the overall performance of the system can be improved. By continuously updating and optimizing the aggregated model, the accuracy and robustness of PCDN service identification can be improved. During the entire PCDN service identification global model construction process, the local node only communicates with the CDN server. In addition to the optimized model parameters maintained together, the local node cannot obtain any information of other local nodes in the CDN in its own region, which ensures the confidentiality of the privacy data of multiple local nodes. By training the received PCDN service identification global model based on the PCDN service identification personalized feature data of the local node, a PCDN service identification personalized model adapted to localization is finally obtained, thereby improving the accuracy of PCDN service identification.

[0084] Embodiment 2:

[0085] like Figure 3 As shown, this embodiment provides a method for identifying a point-to-point content delivery network PCDN service, which is applied to a CDN server of a content delivery network CDN in a target area.

[0086] In this embodiment, the content distribution network CDN includes multiple areas, and there are several clients, i.e., local nodes, in each area. CDN is a regular edge computing network of the operator, and is a layer of intelligent virtual network built on the basis of the existing Internet. It mainly deploys a CDN server in each CDN area as the central server of its own area. The CDN server is responsible for data distribution and collection of local nodes in its own area. The CDN server collects and analyzes the DPI data and traffic monitoring data of all local nodes in its own area, generates the source station content of the CDN server, and distributes the source station content of the CDN server to all local nodes in its own area, so that users of local nodes can obtain the required content nearby. The source station content includes but is not limited to: analysis result information and storage information of DPI data and traffic monitoring data.

[0087] Methods for identifying PCDN services include:

[0088] S201, jointly establishing a PCDN service identification global model with multiple local nodes based on federated learning.

[0089] In this embodiment, through a centralized federated learning framework, the local business data and centrally collected data of each local node in the target area CDN network are retained in the target area CDN network, and a virtual shared model is established by combining data from multiple participants to implement the PCDN service identification algorithm.

[0090] Before S201: jointly establishing a global model for PCDN service identification based on federated learning and multiple local nodes, the PCDN service identification method further includes steps S203 to S206:

[0091] S203, receiving encrypted local business data sent by multiple local nodes, wherein the local business data is used to represent business data collected by the local nodes.

[0092] In this embodiment, the data involved in federated learning modeling includes local business data collected by each local node. When each local node accumulates its own local business data, it will not only identify the user ID of the business data, but also form a set of local business data distinguished by the user phone number by taking the user ID as the user phone number in this embodiment. When a local node collects its own local business data, it will also mark its own data source identifier for its own local business data. The data source identifier is used to characterize the source node of the local business data.

[0093] S204, obtaining the BSS service data encrypted by the BSS node.

[0094] In this embodiment, the data involved in the federated learning modeling also includes BSS service data collected by the operator's BSS node. Similarly, the BSS service data collected by the operator's BSS node also forms a BSS service data set distinguished by user telephone number.

[0095] S205, aligning the encrypted local service data with the encrypted BSS service data based on the user ID, to obtain PCDN service identification feature data, wherein the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data.

[0096] In this embodiment, after collecting and identifying the local service data of each local node and the BSS service data of the BSS node, in order to avoid leaking data privacy, each local node and the BSS node encrypt their own local service data and BSS service data respectively. In addition, the prior art usually collects the local service data of each local node in a centralized manner, and identifies the service data of the PCDN service users of each local node through DPI traffic identification and traffic anomaly detection methods, such as rule identification and AI identification methods. Therefore, in this embodiment, the DPI traffic identification and traffic anomaly detection methods can also be preset on the CDN server. After each local node sends its own local service data to the CDN server, and the BSS node sends the BSS service data to the CDN server, the CDN server can preliminarily identify the local service data of each local node and the BSS service data of the BSS node, the corresponding service data of the PCDN service users of each local node and the BSS node, and identify the corresponding service data of the PCDN service users of each local node and the BSS node through the preset DPI traffic identification and traffic anomaly detection methods.

[0097] Furthermore, the CDN server can confirm the common users of both parties without exposing their respective business data on the BSS node and all local nodes, and the business data of users that are not exposed or overlapped, so as to combine the business data of the common users and realize the mutual supplementation of local business data and BSS business data under the common users of all local nodes and BSS nodes, so as to obtain the PCDN business identification feature data of all local nodes. Specifically, the CDN server adopts the Paillier homomorphic encryption algorithm to align and fuse the encrypted local business data and BSS business data under the same user ID of all local nodes and BSS nodes, so as to obtain the PCDN business identification feature data of all local nodes.

[0098] Based on the data source identifier of the local business data in the PCDN business identification feature data, the PCDN business identification feature data of all local nodes are divided into the PCDN business identification feature data of each local node, and the PCDN business identification feature data of each local node are distributed to each local node. For different local nodes in the same regional CDN, the characteristics of the PCDN business identification feature data are that there is a lot of feature overlap and little user overlap. Therefore, before dividing the PCDN business identification feature data of all local nodes into the PCDN business identification feature data of each local node, this embodiment also aggregates the PCDN business identification feature data of all local nodes under the same regional CDN, determines the PCDN business identification common feature data and PCDN business identification personalized feature data of all local nodes under the same regional CDN, and uses the PCDN business identification common feature data of all local nodes under the same regional CDN to train the PCDN business identification global model, which can maximize the number of samples and improve the accuracy of the global model.

[0099] S206: Send the PCDN service identification feature data to each local node, so that each local node uses the PCDN service identification common feature data as training data for establishing a PCDN service identification global model to train a local PCDN service identification model.

[0100] In this embodiment, after all local nodes obtain their own common feature data for PCDN service identification, the common feature data for PCDN service identification of all local nodes under the same regional CDN is used to establish and train a global model for PCDN service identification under the current regional CDN, so as to obtain a global model for PCDN service identification of the current regional CDN with strong generalization ability. In the case that there are many local nodes under the current regional CDN and the network traffic of the local nodes is large, the overall trend and characteristics of the feature data for PCDN service identification of all local nodes under the current regional CDN can be captured, and a global model for PCDN service identification can be quickly built.

[0101] The PCDN service identification model refers to the global model initially constructed by the CDN server based on the classification algorithm. The CDN server distributes the initially constructed global model to all local nodes of the CDN in the current area, so that each local node trains the initially constructed global model separately to obtain a local PCDN service identification model.

[0102] S202, sending the PCDN service identification global model to each local node, so that each local node superimposes the PCDN service identification personalized feature data to train the PCDN service identification global model, obtains the PCDN service identification personalized model, and identifies its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0103] In this embodiment, considering the differences between BSS service data and local service data, and the different local service data of local nodes under each regional CDN, in the federated learning architecture, each regional CDN adopts the same training model structure, which obviously affects the accuracy of the model. Therefore, in this embodiment, a global PCDN service identification model that is adapted to all local nodes under each regional CDN is not constructed from the BSS node globally, but all local nodes under the same regional CDN perform training on the global PCDN service identification model.

[0104] In addition, since the PCDN service identification common feature data of all local nodes under each regional CDN has different data volume scales and different overall distributions of data features, the contribution of the PCDN service identification common features of different local nodes to the global model is also different. Therefore, this embodiment trains the local PCDN service identification model using the PCDN service identification common feature data at the local node, and then aggregates the model parameters of multiple local nodes based on federated learning through the CDN server to generate a PCDN service identification global model that meets the convergence state.

[0105] While the local node uses the common feature data of PCDN service identification to train the local PCDN service identification model, the local node will calculate the model parameters of its own local PCDN service identification model and send the model parameters to the CDN server of the CDN in the area where the local node is located. The model parameters include but are not limited to: gradient information, statistical information, encryption information, metadata, and convergence speed. In this embodiment, the model parameters are the gradient information of the loss function. For example, the gradient information of the loss function Used to characterize the adjustment requirement update of the model parameter ω when minimizing the loss function.

[0106] The CDN server of the CDN in the area where the local node is located receives the model parameters of all local nodes of the CDN in the area where the local node is located to construct a global model for PCDN service identification of the CDN in the area where the local node is located, specifically: ① weighted sum of the model parameters of all local nodes of the CDN in the area where the local node is located, and calculate the optimized model parameters of all local nodes of the CDN in all areas of the local node; ② distribute the optimized model parameters of all local nodes to all local nodes of the CDN in the area where the local node is located to update the local PCDN identification model of all local nodes; ③ for the model parameters obtained in different rounds, repeat ① and ② until the local PCDN identification model of all local nodes meets the convergence state, or reaches the threshold of the number of repeated executions, aggregate the model parameters of all local nodes of the CDN in the area where the local node is located, and generate a global model for PCDN service identification. In this embodiment, during the entire PCDN service identification global model construction process, the local node only communicates with the CDN server. In addition to the jointly maintained optimized model parameters, the local node cannot obtain any information of other local nodes of the CDN in its own area, thereby ensuring the confidentiality of the privacy data of multiple local nodes.

[0107] Calculate the optimization model parameters of all local nodes of all CDNs in the local node region, as shown in formula (1): Where N represents the total number of local nodes, n j represents the sample weight of the jth local node, n i Represents the sample weight of the i-th local node. Aggregate the model parameters of all local nodes of the CDN in the area where the local node is located to generate a global model for PCDN service identification, as shown in formula (2): Where F(θ) service represents the global model of PCDN service identification of the CDN in the region where the local node is located. The CDN in the region where the local node is located includes K local nodes. k (θ) represents the local PCDN service identification model of the kth local node, n k represents the sample weight of the kth local node, and n represents the sum of the sample weights of the K local nodes. Therefore, the model parameters of the PCDN service identification global model are obtained by weighted summing the model parameters of each local PCDN service identification model according to the sample weight values.

[0108] Each local node will superimpose its own PCDN service identification common feature data based on its own PCDN service identification personalized feature data to train the received PCDN service identification global model. That is, based on the training data of the PCDN service identification global model of each local node including its own PCDN service identification common feature data, its own PCDN service identification personalized feature data is also connected to the training data of the PCDN service identification global model for training the PCDN service identification global model, and finally a PCDN service identification personalized model adapted to localization is obtained.

[0109] The present embodiment provides a method for identifying PCDN services. A CDN server combines common feature data for identifying PCDN services of multiple local nodes of a target area CDN to quickly establish a global model for identifying PCDN services with strong generalization capability, and distributes the model to multiple local nodes of the target area CDN. The global model for identifying PCDN services is then trained using the personalized feature data for identifying PCDN services of each local node of the target area CDN, and a personalized model for identifying PCDN services that meets the network traffic identification requirements of the local node itself is generated. The personalized model for identifying PCDN services can not only quickly analyze and process common feature data for identifying PCDN services, but can also quickly analyze and process personalized feature data for identifying PCDN services, and can then quickly and accurately identify its own PCDN service users, thereby achieving efficient and accurate PCDN service identification, improving the data analysis capability of the PCDN service identification model, improving the accuracy and efficiency of analyzing and processing common feature data and personalized feature data, and reducing maintenance and update costs while achieving more accurate and personalized PCDN service identification services.

[0110] Embodiment 3:

[0111] like Figure 4 As shown, this embodiment further provides a method for identifying a PCDN service, and the method for identifying a PCDN service includes:

[0112] S1, each local node under the CDN in the target area collects its own local business data, and the CDN server obtains the BSS business data of the BSS node.

[0113] In this embodiment, Figure 4 The data preparation in the process is to obtain the local service data of each local node and the BSS service data of the BSS node.

[0114] S2, the CDN server matches and aligns the local service data and BSS service data collected by each local node.

[0115] In this embodiment, Figure 4The local data in the data is the local business data collected by each local node. Figure 4 The operator BSS data in it is BSS service data.

[0116] S3, based on federated learning, the CDN server and multiple local nodes jointly establish a global model for PCDN service identification.

[0117] In this embodiment, Figure 4 The global model in is the PCDN service identification global model.

[0118] S4, each local node under the target area CDN trains the PCDN service identification global model to obtain a PCDN service identification personalized model.

[0119] In this embodiment, Figure 4 The personalized model in is the PCDN service identification personalized model.

[0120] The present embodiment provides a method for identifying PCDN services. A CDN server combines common feature data for identifying PCDN services of multiple local nodes of a target area CDN to quickly establish a global model for identifying PCDN services with strong generalization capability, and distributes the model to multiple local nodes of the target area CDN. The global model for identifying PCDN services is then trained using the personalized feature data for identifying PCDN services of each local node of the target area CDN, and a personalized model for identifying PCDN services that meets the network traffic identification requirements of the local node itself is generated. The personalized model for identifying PCDN services can not only quickly analyze and process common feature data for identifying PCDN services, but can also quickly analyze and process personalized feature data for identifying PCDN services, and can then quickly and accurately identify its own PCDN service users, thereby achieving efficient and accurate PCDN service identification, improving the data analysis capability of the PCDN service identification model, improving the accuracy and efficiency of analyzing and processing common feature data and personalized feature data, and reducing maintenance and update costs while achieving more accurate and personalized PCDN service identification services.

[0121] Embodiment 4:

[0122] like Figure 5As shown, this embodiment also provides a local node, which is applied to the target area CDN, including: a first establishment module 41, a training module 42 and an identification module 43. The first establishment module 41 is used to jointly establish a PCDN service identification global model with other local nodes in a plurality of local nodes through a CDN server based on federated learning. The training module 42 is connected to the first establishment module 41 and is used to superimpose service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model. The identification module 43 is connected to the training module 42 and is used to identify its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0123] Optionally, the local node also includes: a first sending module 44, a first receiving module 45 and a first determining module 46, the first sending module 44 is used to encrypt the local business data collected by itself and send it to the CDN server, the first receiving module 45 is used to receive PCDN business identification feature data sent by the CDN server, wherein the PCDN business identification feature data is obtained by the CDN server by aligning the encrypted local business data with the BSS business data encrypted by the BSS node based on user ID matching, the PCDN business identification feature data includes PCDN business identification common feature data and PCDN business identification personalized feature data, and the first determining module 46 is used to use the PCDN business identification common feature data as training data for establishing a PCDN business identification global model.

[0124] Optionally, the local node also includes: a collection module 47 for collecting local business data, wherein the local business data includes a user ID and at least one of the following: user business data, user signaling data, user business frequency, average daily uplink and downlink traffic of the business, business flow direction, and routing information.

[0125] Specifically, the first establishment module 41 includes: a training unit 411, a first sending unit 412 and a first receiving unit 413. The training unit 411 is used to train the PCDN service identification model based on the common feature data of PCDN service identification to obtain the local PCDN service identification model. The first sending unit 412 is used to send the model parameters of the trained local PCDN service identification model to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, where N is a positive integer. The first receiving unit 413 is used to receive the PCDN service identification global model sent by the CDN server.

[0126] Specifically, the training module 42 includes: a second sending unit 421, a second receiving unit 422, a judging unit 423 and a third sending unit 424, the second sending unit 421 is used to send the model parameters of the local PCDN service identification model after the i-th round of training to the CDN server, so that the CDN server performs weighted summation on the model parameters of multiple local nodes received in the i-th round to generate the optimized model parameters of the i-th round, wherein i=1, 2, ..., N, N represents the round threshold, the second receiving unit 422 is used to receive the i-th round optimized model parameters sent by the CDN server, and adjust The local PCDN service identification model after the i-th round of training is integrated to obtain the local PCDN service identification model after the i+1-th round of training, the judging unit 423 is used to judge whether the local PCDN service identification model after the i+1-th round of training meets the convergence state, and the third sending unit 424 is used to send the model parameters of the local PCDN service identification model after the i+1-th round of training to the CDN server in response to the local PCDN service identification model after the i+1-th round of training meeting the convergence state, so that the CDN server aggregates the model parameters of multiple local nodes to generate a PCDN service identification global model.

[0127] It can be understood that the local node provided above executes the PCDN service identification method corresponding to the embodiment 1 provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the scheme corresponding to the PCDN service identification method of the embodiment 1 above, and will not be repeated here.

[0128] Embodiment 5:

[0129] like Figure 6 As shown, this embodiment also provides a CDN server, which is applied to the target area CDN, and includes a second establishment module 51 and a second sending module 52. The second establishment module 51 is used to jointly establish a PCDN service identification global model with multiple local nodes based on federated learning. The second sending module 52 is connected to the second establishment module 51 and is used to send the PCDN service identification global model to each local node, so that each local node superimposes the PCDN service identification personalized feature data to train the PCDN service identification global model to obtain the PCDN service identification personalized model, and identifies its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

[0130] Optionally, the CDN server also includes: a second receiving module 53, an acquisition module 54, an alignment module 55 and a third sending module 56, the second receiving module 53 is used to receive encrypted local business data sent by multiple local nodes, wherein the local business data is used to characterize the business data collected by the local node, the acquisition module 54 is used to obtain the BSS business data encrypted by the BSS node, the alignment module 55 is used to align the encrypted local business data with the encrypted BSS business data based on user ID matching to obtain PCDN business identification feature data, wherein the PCDN business identification feature data includes PCDN business identification common feature data and PCDN business identification personalized feature data, and the third sending module 56 is used to send the PCDN business identification feature data to each local node, so that each local node uses the PCDN business identification common feature data as training data for establishing a PCDN business identification global model to train a local PCDN business identification model.

[0131] It can be understood that the CDN server provided above executes the PCDN service identification method corresponding to the embodiment 2 provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the scheme corresponding to the PCDN service identification method of the embodiment 2 above, which will not be repeated here.

[0132] Embodiment 6:

[0133] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the PCDN service identification method in the above-mentioned embodiment 1 or embodiment 2.

[0134] Embodiment 7:

[0135] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for identifying the PCDN service in the above-mentioned embodiment 1 or embodiment 2 is implemented.

[0136] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying a point-to-point content delivery network (PCDN) service, applied to any local node among a plurality of local nodes of a target area content delivery network (CDN), characterized in that: Methods for identifying PCDN services include: Based on federated learning, a global model for PCDN service identification is established by combining the CDN server with other local nodes in multiple local nodes; Superimposing the PCDN service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model; The network traffic is identified based on the PCDN service identification personalized model to identify the PCDN service user.

2. The method for identifying a PCDN service according to claim 1, characterized in that: Before the method of establishing a PCDN service identification global model through the CDN server and other local nodes in the plurality of local nodes based on federated learning, the method further includes: Encrypt the local business data collected by itself and send it to the CDN server; Receiving PCDN service identification feature data sent by the CDN server, wherein the PCDN service identification feature data is obtained by the CDN server by aligning the encrypted local service data with the BSS service data encrypted by the BSS node based on the user ID matching, and the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data; The PCDN service identification common feature data is used as training data for establishing a PCDN service identification global model.

3. The method for identifying a PCDN service according to claim 1, characterized in that: The method of establishing a PCDN service identification global model through the CDN server and other local nodes in the multiple local nodes based on federated learning specifically includes: Based on the PCDN service identification common feature data, a PCDN service identification model is trained to obtain a local PCDN service identification model; Sending the model parameters of the trained local PCDN service identification model to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, where N is a positive integer; Receive the PCDN service identification global model sent by the CDN server.

4. The method for identifying a PCDN service according to claim 3, characterized in that: The sending of the model parameters of the trained local PCDN service identification model to the CDN server so that the CDN server aggregates the model parameters of multiple nodes received in N rounds based on federated learning to generate a PCDN service identification global model that meets the convergence state, specifically includes: Sending the model parameters of the local PCDN service identification model after the i-th round of training to the CDN server, so that the CDN server performs weighted summation on the model parameters of multiple local nodes received in the i-th round to generate the optimized model parameters of the i-th round, where i=1, 2, ..., N, and N represents the round threshold; Receiving the optimization model parameters of the i-th round sent by the CDN server, adjusting the local PCDN service identification model after the i-th round of training, and obtaining the local PCDN service identification model after the (i+1)-th round of training; Determine whether the local PCDN service identification model after the i+1th round of training meets the convergence state; In response to the local PCDN service identification model after the i+1th round of training meeting the convergence state, the model parameters of the local PCDN service identification model after the i+1th round of training are sent to the CDN server, so that the CDN server aggregates the model parameters of multiple local nodes to generate a PCDN service identification global model.

5. The method for identifying a PCDN service according to claim 2, characterized in that: BSS service data includes user ID and at least one of the following: user real-name registration information, user location data, package type, consumption amount information, and location information. Before encrypting the local business data collected by itself and sending it to the CDN server, it also includes: Collect local business data, wherein the local business data includes a user ID and at least one of the following: user business data, user signaling data, user business frequency, daily average uplink and downlink traffic of the business, business flow direction, and routing information.

6. A method for identifying a point-to-point content delivery network (PCDN) service, applied to a CDN server of a target area content delivery network (CDN), characterized in that: Methods for identifying PCDN services include: Based on federated learning, a global model for PCDN service identification is established in conjunction with multiple local nodes. The PCDN service identification global model is sent to each local node, so that each local node superimposes the PCDN service identification personalized feature data to train the PCDN service identification global model, obtains the PCDN service identification personalized model, and identifies its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

7. The method for identifying a PCDN service according to claim 6, characterized in that: Before the method of jointly establishing a PCDN service identification global model with multiple local nodes based on federated learning, the method further includes: Receiving encrypted local business data sent by multiple local nodes, wherein the local business data is used to represent business data collected by the local nodes; Obtain BSS service data encrypted by the BSS node; Align the encrypted local service data with the encrypted BSS service data based on the user ID to obtain PCDN service identification feature data, wherein the PCDN service identification feature data includes PCDN service identification common feature data and PCDN service identification personalized feature data; The PCDN service identification feature data is sent to each local node, so that each local node uses the PCDN service identification common feature data as training data for establishing a PCDN service identification global model to train a local PCDN service identification model.

8. A local node, applied to a target area CDN, characterized in that: It includes a first establishment module, a training module and a recognition module, The first establishment module is used to establish a PCDN service identification global model through the CDN server and other local nodes in the multiple local nodes based on federated learning, A training module, connected to the first establishment module, is used to superimpose the service identification personalized feature data to train the PCDN service identification global model to obtain a PCDN service identification personalized model. The identification module is connected to the training module and is used to identify its own network traffic based on the PCDN service identification personalized model to identify PCDN service users.

9. A CDN server, applied to a target area CDN, characterized in that: comprising a second establishing module and a second sending module, The second establishment module is used to jointly establish a PCDN service identification global model with multiple local nodes based on federated learning. The second sending module is connected to the second establishing module, and is used to send the PCDN service identification global model to each local node, so that each local node superimposes the PCDN service identification personalized feature data to train the PCDN service identification global model, obtains the PCDN service identification personalized model, and identifies its own network traffic based on the PCDN service identification personalized model to identify the PCDN service user.

10. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the method for identifying a PCDN service as claimed in any one of claims 1 to 5 or the method for identifying a PCDN service as claimed in any one of claims 6 to 7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying a PCDN service as described in any one of claims 1 to 5 or the method for identifying a PCDN service as described in any one of claims 6 to 7 is implemented.

Images