CN-121985157-A - Point-to-point content distribution network flow identification method using storage memory prototype network

CN121985157ACN 121985157 ACN121985157 ACN 121985157ACN-121985157-A

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a point-to-point content distribution network flow identification method for a storage memory prototype network, which comprises the steps of collecting flow samples, establishing an original data set, constructing a plurality of small sample learning tasks, wherein each small sample learning task comprises a plurality of initial categories, and respectively defining a support set and a query set for any small sample learning task; the method comprises the steps of constructing a feature extraction network, adopting a convolution architecture with a depth self-adaptive adjustment mechanism, carrying out initialization configuration on network parameters, respectively inputting a support set and a query set into the initialized feature extraction network to generate corresponding embedded feature representations, training and optimizing the network parameters by adopting the embedded feature representations, constructing a category prototype for an initial category corresponding to the support set based on the embedded feature representations, storing the category prototype to establish a prototype memory module, analyzing the embedded feature representations corresponding to the query set by using the prototype memory module, judging the category of a flow sample, and generating a recognition result.

Inventors

ZHOU PENG
JIANG YUNA
LI ZEYI

Assignees

南京邮电大学

Dates

Publication Date: 20260505
Application Date: 20260313

Claims (10)

1. A method for identifying traffic in a point-to-point content delivery network for a storage memory prototype network, the method comprising: collecting flow samples, establishing an original data set, and establishing a plurality of small sample learning tasks based on the original data set, wherein each small sample learning task comprises a plurality of initial categories, and a support set and a query set are respectively defined for any small sample learning task; constructing a feature extraction network, adopting a convolution architecture with a depth self-adaptive adjustment mechanism, and carrying out initialization configuration on network parameters of the convolution architecture; Respectively inputting the support set and the query set into the initialized feature extraction network to generate corresponding embedded feature representations, and training and optimizing network parameters by adopting the embedded feature representations; building a class prototype for the initial class corresponding to the support set based on the embedded feature representation, and storing the class prototype to build a prototype memory module; And analyzing the embedded characteristic representation corresponding to the query set by using the prototype memory module, judging the class of the flow sample of the query set, and generating an identification result.
2. The method of claim 1, wherein the number of traffic samples used to construct the support set in each initial class is the same in the same small sample learning task.
3. The method for identifying the flow of the point-to-point content distribution network for storing and memorizing a prototype network according to claim 1, wherein the feature extraction network comprises a plurality of cascaded one-dimensional convolution blocks, a time sequence feature extraction module and a feature mapping module at the tail end, wherein the one-dimensional convolution blocks comprise a one-dimensional convolution layer, a one-dimensional batch normalization layer, a ReLU activation layer and a one-dimensional maximum pooling layer which are sequentially arranged, the time sequence feature extraction module adopts a two-way gating circulation unit or a two-way long-short-term memorizing network, and the feature mapping module comprises at least three full-connection layers.
4. The method for identifying peer-to-peer content distribution network traffic in a storage and retrieval prototype network according to claim 3, wherein the network operation mode corresponding to the feature extraction network is as follows: ; Wherein, the Represent the first Embedded feature representations corresponding to the individual flow samples; a mapping function representing a feature extraction network, A trainable parameter set representing a feature extraction network; represent the first And a number of traffic samples.
5. The method for identifying traffic of a point-to-point content distribution network for a storage and retrieval prototype network according to claim 4, wherein the steps of inputting a support set and a query set into the initialized feature extraction network, respectively, generating a corresponding embedded feature representation, training and optimizing network parameters using the embedded feature representation, and comprising: respectively obtaining embedded feature representations corresponding to the support set and the query set based on the initialized feature extraction network; And adopting feature embedding representation to sequentially perform discriminant pre-training and prototype constraint optimization on the feature extraction network, constructing a cross entropy loss function in the discriminant pre-training stage to obtain basic discriminant force, and constructing a prototype loss function in the prototype constraint optimization stage to refine embedded space distribution, so as to update network parameters step by step and iteration.
6. The method for identifying traffic in a peer-to-peer content distribution network for a storage and retrieval prototype network according to claim 5, wherein the performing the discriminant pre-training and the prototype constraint optimization on the feature extraction network in sequence using the feature embedding representation, and the correspondingly constructing the cross entropy loss function and the prototype loss function, updating the network parameters, comprises: Connecting a classification layer after the embedded feature representation, predicting the class of the flow sample to obtain a prediction result, constructing a cross entropy loss function, and updating network parameters by minimizing the cross entropy loss function; and freezing the parameters of the convolution layer in the feature extraction network, only updating the network parameters of the full connection layer, constructing a prototype loss function, and optimizing the spatial structure of the embedded feature representation by minimizing the prototype loss function.
7. The method for identifying traffic of a peer-to-peer content distribution network for a storage and retrieval prototyping network of claim 1 wherein building a class prototype for an initial class corresponding to a support set based on embedded feature representation and storing the class prototype building prototyping memory module comprises: Generating a corresponding category prototype for each initial category corresponding to the support set through the embedded feature representation, defining the category prototype as a representative feature of the embedded feature representation, storing the representative feature to establish a prototype memory module, fusing the representative feature corresponding to the initial category in the small sample learning task according to the current analysis with the stored historical representative feature, and updating the category prototype in real time.
8. The method for identifying traffic of a peer-to-peer content distribution network for storing and memorizing a prototype network according to claim 7, wherein analyzing the embedded feature representation corresponding to the query set by using the prototype memory module, performing class judgment on the traffic sample of the query set, and generating an identification result, comprises: performing anomaly detection pre-screening based on the flow samples in the query set, judging whether the flow samples belong to the baseline normal distribution, if so, triggering the forced coverage mechanism to directly generate the identification result, and if not, performing the subsequent judging step; When the flow sample is judged to be abnormal flow, obtaining a distance relation between the flow sample and each prototype type in the prototype memory module based on the embedded feature representation corresponding to the query set; And judging the prediction type of the flow sample in the query set through the distance relation, and generating a recognition result.
9. The method for identifying traffic of a peer-to-peer content distribution network for storing and memorizing a prototype network according to claim 8, wherein a class probability distribution is constructed based on a distance relation, and a fusion ratio of representative features corresponding to a query set and stored historical representative features of the prototype memory module is adjusted for improving stability of a predicted class.
10. The electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus, and the processor calls logic instructions in the memory to execute the point-to-point content distribution network traffic identification method for storing a memory prototype network according to any one of claims 1-9.

Description

Point-to-point content distribution network flow identification method using storage memory prototype network Technical Field The invention relates to the technical field of artificial intelligence, in particular to a point-to-point content distribution network traffic identification method for a storage memory prototype network. Background In recent years, video content providers have begun to attempt to employ a new type of content delivery system, a point-to-Point Content Distribution Network (PCDN), or Peer-to-Peer Content Delivery Network, to build low-cost, high-quality network services. However, the PCDN system gradually exposes a significant resource imbalance problem due to the expansion of service scale, the continuous increase of audience quantity and the continuous introduction of heterogeneous edge nodes, and part of edge nodes are in an overload state for a long time, so that the overall efficiency of the system is severely restricted. In particular, in a real network environment, PCDN traffic is often highly mixed with other application traffic, and is significantly affected by terminal behavior, node heterogeneity and network state fluctuation, so that it is difficult for the network side to form an observable and interpretable fine representation of "who is transmitting, what is transmitting, how much is transmitting, and when. Under the condition of lacking reliable flow identification capability, the network management system is difficult to carry out differentiated bandwidth guarantee, congestion control and resource scheduling, so that the performance degradation of the overload node and the idle problem of the low-cost node are further amplified. Disclosure of Invention In order to solve the technical problems that in the existing network environment, traffic data cannot be accurately identified and described, and network management and service optimization are affected, the invention aims to provide a point-to-point content distribution network traffic identification method for storing and memorizing a prototype network, and the adopted technical scheme is as follows: collecting flow samples, establishing an original data set, and establishing a plurality of small sample learning tasks based on the original data set, wherein each small sample learning task comprises a plurality of initial categories, and a support set and a query set are respectively defined for any small sample learning task; constructing a feature extraction network, adopting a convolution architecture with a depth self-adaptive adjustment mechanism, and carrying out initialization configuration on network parameters of the convolution architecture; Respectively inputting the support set and the query set into the initialized feature extraction network to generate corresponding embedded feature representations, and training and optimizing network parameters by adopting the embedded feature representations; building a class prototype for the initial class corresponding to the support set based on the embedded feature representation, and storing the class prototype to build a prototype memory module; And analyzing the embedded characteristic representation corresponding to the query set by using the prototype memory module, judging the class of the flow sample of the query set, and generating an identification result. Preferably, the number of traffic samples used to construct the support set within each initial class is consistent in the same small sample learning task. The feature extraction network comprises a plurality of cascaded one-dimensional convolution blocks, a time sequence feature extraction module and a feature mapping module at the tail end, wherein the one-dimensional convolution blocks comprise a one-dimensional convolution layer, a one-dimensional batch normalization layer, a ReLU activation layer and a one-dimensional maximum pooling layer which are sequentially arranged, the time sequence feature extraction module adopts a two-way gating circulation unit or a two-way long-short-term memory network, and the feature mapping module comprises at least three full-connection layers. Preferably, the network operation mode corresponding to the feature extraction network is as follows: Wherein, the Represent the firstEmbedded feature representations corresponding to the individual flow samples; a mapping function representing a feature extraction network, A trainable parameter set representing a feature extraction network; represent the first And a number of traffic samples. Preferably, the support set and the query set are respectively input into the initialized feature extraction network to generate corresponding embedded feature representations, and the embedded feature representations are adopted to train and optimize network parameters, including: respectively obtaining embedded feature representations corresponding to the support set and the query set based on the initialized feature extraction netwo