CN113660114A

CN113660114A - Reconstruction method, system and medium for distributed network random space sampling measurement

Info

Publication number: CN113660114A
Application number: CN202110854263.9A
Authority: CN
Inventors: 谢逸
Original assignee: Sun Yat Sen University
Current assignee: Sun Yat Sen University
Priority date: 2021-07-28
Filing date: 2021-07-28
Publication date: 2021-11-16
Anticipated expiration: 2041-07-28
Also published as: CN113660114B

Abstract

The invention discloses a reconstruction method, a system and a medium for random space sampling measurement of a distributed network, wherein the method comprises the following steps: converting the network topology into a hypergraph structure; constructing a multi-layer hypergraph according to the network topology under the hypergraph structure; the multi-layer hypergraph is used for representing a large-scale distributed network measurement result; training to obtain a reconstructed model according to the multi-layer hypergraph; and restoring the result obtained by incomplete measurement in the actual measurement service into complete network measurement data according to the reconstruction model. The invention improves the measurement efficiency and reduces the measurement cost, and can be widely applied to the technical field of network measurement.

Description

Reconstruction method, system and medium for distributed network random space sampling measurement

Technical Field

The invention relates to the technical field of network measurement, in particular to a reconstruction method, a reconstruction system and a reconstruction medium for random space sampling measurement of a distributed network.

Background

The continuous development and evolution of the network promote the continuous increase of the variety and scale of the network and the increasingly prominent heterogeneity of the network, and a plurality of network systems with different properties and different structures and functions emerge to form the situation of coexistence of various heterogeneous networks. These networks include: the system comprises IPv4 and IPv6 packet networks, a telephone telecommunication network, 2G/3G/4G-LTE/5G and other wireless network systems, a cloud computing-oriented data center network, an intelligent terminal-oriented Internet of things, an industrial control network and an air-space-sea-ground integrated network. Emerging network architectures, such as: multiprotocol label networks, software defined networks, content centric networks, etc. have also gained widespread attention and use. On the other hand, modern network services not only realize the fusion of traditional television, telephone and data, but also increase the types, scales and heterogeneity of services. The rapid development of network services has migrated a large amount of services in the physical world into virtual networks, such as: network medical services, e-finance and commerce, government affairs, online education and office and various services, and the like.

The rapid development of network infrastructure and data services has prompted the overlapping transmission of large amounts of heterogeneous service data over large-scale heterogeneous communication networks, resulting in a large number of new problems and challenges for network management, including: route optimization, load balancing, service customization, security management and control and the like. The accurate control of the working state of the communication network is a key link for realizing efficient network management and is an important means for ensuring the normal operation of the network.

Network measurement is the basis and foundation for network managers to obtain network state information and implement network management. As can be seen from the published technical literature and data, most of the existing network measurement work is mainly focused on local fixed points, and is not suitable for the distributed network scenario. The limitation of the prior art is the lack of systematic measurements towards distributed networks. The small-scale network-oriented full-measurement method is not suitable for a large-scale distributed network scenario, and may cause a series of problems, for example: increased measurement cost, increased network load, increased measurement delay, difficulty in synchronizing measurement clocks, etc. Spatial sampling measurement based on network topology is an optional compromise, but the main problem is that spatial sampling causes data loss, which has a great influence on later data analysis. To solve this problem, some practical distributed network measurements usually adopt a static sampling measurement method, that is, measurement equipment is deployed at multiple fixed points in the network to realize spatial sampling. Under the static sampling measurement mode, the context relationship between the default point and the measured point is fixed, so that the data recovery is facilitated. However, it cannot adapt to dynamically changing network topology, and static sampling can result in signals at the points of absence never being measured, thereby possibly losing important information. In addition, static sampling measurements are vulnerable to attack and hijacking, affecting the availability of measurement data. Random spatial sampling can dynamically cope with a changeable network structure, and more comprehensive data can be acquired by changing sampling points. However, the main problems of the method are that in a large-scale, distributed and irregular network scene, the random space sampling causes uncertain relation between a defect point and a measured point, and the dynamic property of the sampling makes data difficult to recover, thereby affecting the usability of the data.

Disclosure of Invention

In view of this, embodiments of the present invention provide a reconstruction method, system and medium for random spatial sampling measurement in a distributed network, so as to improve measurement efficiency and reduce measurement cost.

One aspect of the present invention provides a reconstruction method for random spatial sampling measurement of a distributed network, including:

converting the network topology into a hypergraph structure;

constructing a multi-layer hypergraph according to the network topology under the hypergraph structure; the multi-layer hypergraph is used for representing a large-scale distributed network measurement result;

training to obtain a reconstructed model according to the multi-layer hypergraph;

and restoring the result obtained by incomplete measurement in the actual measurement service into complete network measurement data according to the reconstruction model.

Optionally, the converting the network topology into a hypergraph structure includes:

respectively endowing nodes in the distributed network topology with a unique ID identifier and carrying out one-hot coding;

extracting an end-to-end path sequence set in the distributed network;

cutting each end-to-end path sample in the path sequence set into equal length path segments;

projecting the one-hot coded network node to a coordinate point in a high-dimensional vector space according to the cut path segment, and expressing the node in the distributed network topology through the coordinate vector of the coordinate point to realize node vectorization;

converting the distributed network topology into a hypergraph formed by stretching the high-dimensional vector space according to the result of the node vectorization; wherein the node in each network topology corresponds to a fixed pixel point in the hypergraph.

Optionally, the constructing a multi-layer hypergraph according to the network topology under the hypergraph structure includes:

constructing a plurality of layers for the network topology of the hypergraph structure; each layer corresponds to a measurement index or a measurement variable on a network element in the distributed network;

the measurement data of a certain index of a network element node on the distributed network is mapped on the gray value of the corresponding layer on the corresponding pixel point on the hypergraph;

the positions of the pixel points of the hypergraph represent the connection and structure information of the network element nodes in the network topology;

the gray value of the pixel point on each layer reflects the measurement data of the network element node.

Optionally, the training according to the multi-layer hypergraph to obtain a reconstructed model includes:

converting a result obtained by incomplete sampling measurement of the distributed network into a hypergraph form, and marking pixels of the missing points through a mask layer;

inputting the hypergraph obtained by conversion and the mask image layer into a trained reconstruction model to obtain gray level recovery values of the defect points on different image layers;

filling the gray restoration value restored from the missing point into the pixel position of the corresponding layer in the incomplete hypergraph to complete the restoration of the missing point and obtain a pseudo hypergraph;

training by utilizing the pseudo hypergraph and the real complete hypergraph to obtain a discriminator, wherein the discriminator is used for distinguishing the pseudo hypergraph from the real complete hypergraph;

and judging the pseudo hypergraph according to the discriminator to determine a real complete hypergraph, wherein when the discriminator cannot accurately identify the true hypergraph and the false hypergraph, the predictor representing the model is trained.

Optionally, the restoring, according to the reconstruction model, a result obtained by incomplete measurement in an actual measurement service to complete network measurement data includes:

filling the gray restoration value restored from the missing point into the pixel position of the corresponding layer in the incomplete hypergraph to finish the restoration of the missing point;

and restoring the recovery result in the hypergraph form into the original network element characteristic form.

The embodiment of the invention also provides a reconstruction system for distributed network random space sampling measurement, which comprises a training subsystem and a reconstruction subsystem, wherein the training subsystem comprises a topological hypergraph expression module, a network measurement hypergraph module and a measurement reconstruction model training module:

the topological hypergraph expression module is used for converting the network topology into a hypergraph structure;

the hypergraphization module for network measurement is used for constructing a multi-layer hypergraph according to the network topology under the hypergraph structure; the multi-layer hypergraph is used for representing a large-scale distributed network measurement result;

the measurement reconstruction model training module is used for training to obtain a reconstruction model according to the multi-layer hypergraph;

and the reconstruction subsystem is used for restoring the result obtained by incomplete measurement in the actual measurement service into complete network measurement data according to the reconstruction model.

Optionally, the hypergraph expression module of the topology includes:

a one-hot coding unit, configured to assign a unique ID identifier to each node in the distributed network topology and perform one-hot coding;

an end-to-end path sampling unit, which is used for extracting an end-to-end path sequence set in the distributed network;

a K-element sub-path splitting unit, configured to split each end-to-end path sample in the path sequence set into equal-length path segments;

the node vectorization unit is used for projecting the one-hot coded network node to a coordinate point in a high-dimensional vector space according to the path segment obtained by cutting, and expressing the node in the distributed network topology through the coordinate vector of the coordinate point so as to realize node vectorization;

the topology hypergraphization unit is used for converting the distributed network topology into a hypergraph formed by stretching the high-dimensional vector space according to the node vectorization result; wherein the node in each network topology corresponds to a fixed pixel point in the hypergraph.

Optionally, the measurement reconstruction model training module includes:

a random mask generator for constructing incomplete measurement samples of the training model;

the hypergraph recovery generator is used for predicting the numerical value of the lack-of-measurement part by utilizing the hypergraph obtained by incomplete measurement to obtain a pseudo hypergraph;

the true-false hypergraph discriminator is used for obtaining a discriminator by utilizing the pseudo hypergraph and the real complete hypergraph training, and the discriminator is used for distinguishing the pseudo hypergraph from the real complete hypergraph; and judging the pseudo-hypergraph according to the discriminator to determine a real complete hypergraph.

Optionally, the reconstruction subsystem comprises:

the measurement reconstruction module is used for converting a result obtained by incomplete sampling measurement of the distributed network into a hypergraph form and marking pixels of the missing points through a mask map layer; inputting the hypergraph obtained by conversion and the mask image layer into a trained reconstruction model to obtain gray level recovery values of the defect points on different image layers; filling the gray restoration value restored from the missing point into the pixel position of the corresponding layer in the incomplete hypergraph to finish the restoration of the missing point;

and the network expression module is used for restoring the recovery result of the hypergraph form into an original network element characteristic form.

Yet another aspect of the embodiments of the present invention provides a computer-readable storage medium, which stores a program, which is executed by a processor to implement the method as described above.

The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.

The embodiment of the invention converts the network topology into a hypergraph structure; constructing a multi-layer hypergraph according to the network topology under the hypergraph structure; the multi-layer hypergraph is used for representing a large-scale distributed network measurement result; training to obtain a reconstructed model according to the multi-layer hypergraph; and restoring the result obtained by incomplete measurement in the actual measurement service into complete network measurement data according to the reconstruction model. The invention improves the measuring efficiency and reduces the measuring cost.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a block diagram of a system according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a block diagram for implementing a hypergraph expression module of a topology according to an embodiment of the present invention;

fig. 3 is a schematic diagram of a network node coding implementation method according to an embodiment of the present invention;

fig. 4 is a schematic diagram illustrating an implementation of a node vectorization method according to an embodiment of the present invention;

fig. 5 is a schematic diagram illustrating an implementation of a measurement reconstruction model training method according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

Aiming at the problems in the prior art, the invention provides a random space sampling measurement reconstruction method and system in a large-scale, distributed and irregular network scene. The method and the system recover the measurement data of the network element lacking the measurement in the network through the designed reconstruction scheme by using the result of the random space measurement in the distributed network. The method and the system are suitable for large-scale, distributed and irregular network scenes, support the data reconstruction of static and dynamic sampling measurement, and can obviously improve various performances of distributed network measurement, including: the method has the advantages of improving the measurement efficiency, reducing the measurement cost, reducing the network traffic, avoiding the measurement point from being hijacked, and the like.

The invention provides a random space sampling measurement reconstruction method and a system facing to a large-scale, distributed and irregular network scene, which comprises the following steps: a training subsystem and a reconstruction subsystem.

The training subsystem is used for establishing a model for random space sampling measurement reconstruction of the distributed network. The input of the training subsystem is a training sample, and the output is a network measurement hypergraph and a reconstructed model parameter. To achieve this, it comprises three functional modules: the system comprises a topological hypergraph expression module, a network measurement hypergraph module and a measurement reconstruction model training module.

The function of the topological hypergraph expression module is to convert the network topology into a hypergraph structure. The input of the module is the topology structure of the distributed network, and the output is the corresponding topology hypergraph. In the present invention, a hypergraph is a regular grid structure resembling a normal planar image. However, unlike a general two-dimensional image, it has a high dimension. The irregular network topology is mapped into the high-dimensional regular hypergraph, so that data recovery can be conveniently carried out by a mathematical method subsequently. To implement this mapping, the hypergraph representation module of the topology comprises 5 sub-modules: one-hot coding, end-to-end path sampling, K-element sub-path segmentation, node vectorization and topology hypergraphization.

The one-hot coding sub-module is used for assigning a unique ID identifier to each node in the large-scale distributed network topology. In order to enable subsequent data processing to have better effect, the invention selects a one-hot coding scheme with the maximum inter-code distance, so that the initial IDs of any two nodes have the same maximum code distance.

The function of the end-to-end path sampling sub-module is to extract the end-to-end path sequence set in the distributed network. The method aims to express the topological structure information of an original network and the spatial connection relation between network elements in the network topology through a complete end-to-end path sample. The input to the module is the distributed network topology and the output is the end-to-end path set. The end-to-end path generation method may be determined according to a specific application scenario, for example: a shortest path method, a weighted path selection method, a random walk method, etc. may be employed. The patent of the present invention does not limit the concrete embodiments.

The function of the K-element sub-path splitting sub-module is to split each obtained end-to-end path sample into equal-length path segments, and each split path segment has a length of K, that is, contains K adjacent network elements. The length K can be selected according to practical requirements, and is generally an odd number. The input of the K-element sub-path division sub-module is the network topology, and the output is the K-element sub-path set. The path segmentation method may be determined according to a specific application scenario, for example: and applying K as a window to each end-to-end path sample, moving according to a specified step length, and extracting a path segment in the window into a new sample in the K-element sub-path set after moving the window once.

The node vectorization sub-module is used for projecting the one-hot coded network node to a coordinate point in a high-dimensional vector space according to the path segment obtained by cutting, and representing the node through a coordinate vector of the coordinate point. The input of the node vectorization sub-module is one-hot coding of each network element in the network topology, and the output is a high-dimensional vector with a specified dimension, which represents the mapping coordinates of the input network element in a high-dimensional vector space. The node vectorization submodule utilizes the neural network to mine and learn topological structure information among network elements from the K-element sub-path sequence set, solidifies the structural information of the network topology into parameters of the neural network through iterative training, and realizes the mapping of the network elements to a high-dimensional vector space according to the topological structure information. The mapping is embodied by topological structure information, which maps network elements which are close to or closely related to each other in the network topology to adjacent areas in a high-dimensional vector space on one hand; on the other hand, the symbolic non-calculable network element ID (one-hot code) is converted into the coordinates of a high-dimensional vector space, so that the network element can perform calculation based on the high-dimensional vector space coordinates, and the subsequent data processing is facilitated.

The topology hypergraph sub-module is used for converting the large-scale distributed network topology into a hypergraph formed by stretching the high-dimensional vector space according to the node vectorization result, so that the conversion from the irregular network topology to the regular hypergraph is realized. The input of the module is one-hot coding of network elements in the distributed network, and the output is a topological hypergraph corresponding to the distributed network. After the vectorization of the nodes is completed, each network element is mapped to a point in a given high-dimensional vector space and has a unique high-dimensional vector space coordinate. All discrete coordinate points in the high-dimensional space form a pixel topology of the high-dimensional hypergraph, and the pixel topology has a regular grid structure. Wherein, the coordinate points mapped by the network elements in the network topology form foreground pixels of the hypergraph, and the coordinate points irrelevant to the network elements in the network topology form background pixels of the hypergraph. To this end, the distributed irregular network topology is converted into a high-dimensional regular hypergraph.

The hypergraphization module for network measurement is used for assigning the measurement result of the nodes in the distributed network to corresponding pixel points in the hypergraphization in a gray scale mode on the basis of topological hypergraphization, so that the multi-layer hypergraph capable of expressing the measurement result of the large-scale distributed network is formed. The input of the hypergraphization module for network measurement is a network measurement result and a topological hypergraph, and the output of the hypergraphization module is a multi-layer hypergraph. Therefore, a hypergraphization module for network measurement firstly constructs a plurality of layers for the obtained topological hypergraph, and each layer corresponds to a measurement index/variable on a network element in a distributed network. Therefore, according to the corresponding relationship between the measurement index and the layer in the hypergraph, the measurement data of a certain index of a network element on the distributed network is mapped to the gray value of the corresponding layer on the corresponding pixel point on the hypergraph. When all network elements finish the gray mapping from the measured data to the super-image layer, the original distributed network measurement result is completely converted into a super-image of a high-dimensional multi-image layer. The positions of the pixel points of the hypergraph represent the connection and structure information of the network element nodes in the network topology, and the gray level of the pixel points on each layer reflects the measurement data of the network element.

The measurement reconstruction model training module is used for training a reconstruction model so that the reconstruction model can reconstruct a hypergraph with complete measurement data according to the hypergraph obtained by random space sampling measurement in the distributed network. It comprises three parts: random mask generator, hypergraph recovery generator, and true and false hypergraph discriminator. The input of the module is a hypergraph of a training sample, and the output is the model parameters of a hypergraph recovery generator.

The random mask generator is used to construct incomplete measurement samples of the training model. The measurement reconstruction model training adopts a supervision mode, so that an incomplete measurement sample is required to be used as the input of the model, and a corresponding complete measurement sample is used as the label of the output end during the model training. The random mask generator is used for generating randomly distributed space mask coordinates for a given complete measurement hypergraph, and pixel point gray values corresponding to the mask coordinates can be eliminated to form a defect point. Thereby obtaining a hypergraph training sample with the missing points. The mask is configured such that a new layer in the hypergraph is input to a subsequent hypergraph restoration generator along with the incomplete measured hypergraph.

The hypergraph recovery generator is used for predicting the numerical value of the lack-of-measurement part of the hypergraph obtained by incomplete measurement to realize the recovery of the measurement data. And the hypergraph recovery generator determines the coordinates of the missing points by using the mask layer, and then encodes the input incomplete hypergraph by using a neural network to fully fuse the missing points and the adjacent tested nodes. And decoding the fused codes by using the neural network again, wherein in the process, the missing point recovers the value of the missing point by using the data of the adjacent nodes. The input of the hypergraph recovery generator is a hypergraph obtained by incomplete measurement and a mask layer marking a missing point, and the output of the hypergraph recovery generator is a recovered complete measurement hypergraph. The invention does not limit the concrete implementation means of the hypergraph recovery generator, and can be selected according to the actual situation, such as: a codec, convolutional neural network, fully-connected depth network, etc. may be used to implement the hypergraph recovery generator.

And replacing the incomplete hypergraph input by the hypergraph recovery generator with the recovery data generated by the hypergraph recovery generator to form a recovered complete hypergraph.

The super-graph discriminator is a supervision and identification model. The invention refers to the complete hypergraph recovered by the hypergraph recovery generator as a pseudo hypergraph. The false hypergraph discriminator firstly trains a discriminator by using the obtained false hypergraph and the real complete hypergraph. The discriminator is a two-classifier in order to correctly distinguish the pseudo-hypergraph from the true full hypergraph. After the training of the discriminator is finished, the discriminator is used for judging the authenticity attribute of the current pseudo-hypergraph. If the result of the discriminator shows that the hypergraph recovery generator generates a fake hypergraph, the restored hypergraph generated by the hypergraph recovery generator does not meet the precision requirement, and the parameters of the hypergraph recovery generator need to be continuously adjusted by using the error of the discriminator, so that the fake hypergraph generated next time can further approach to a real complete hypergraph; if the discriminator judges the current fake hypergraph as a fake hypergraph, the fake hypergraph produced by the generator is close to the real complete hypergraph, and the expected recovery requirement can be met.

The reconstruction subsystem is used for recovering complete network measurement data by using a result obtained by incomplete measurement in actual measurement service based on a recovery model obtained by the training subsystem. It comprises two parts: and measuring the reconstruction and the network expression of the reconstruction result.

The role of the measurement reconstruction model is to recover the incomplete hypergraph obtained from the actual sampling measurement. It consists of a hypergraph recovery generator in the training subsystem. Firstly, converting the result obtained by incomplete sampling measurement of the distributed network into a hypergraph form according to the related steps, and marking the pixels of the missing points through a mask layer. Then, inputting the hypergraph and mask image layer into a trained measurement reconstruction model to obtain the gray level recovery values of the defect points on different image layers. And finally, filling the gray value recovered from the missing point into the pixel position of the corresponding layer in the incomplete hypergraph to finish the recovery of the missing point.

The network expression of the reconstruction result has the function of restoring the recovery result of the hypergraph form into the original network element characteristic form, so that the subsequent data analysis and processing are facilitated. The main basis is the mapping relationship between the network elements and the high-dimensional vector space. When a neural network is used to establish the mapping relation between the network element nodes of the distributed network and the high-dimensional vector space in the training stage, a mapping table can be obtained, wherein the mapping table comprises the one-to-one correspondence relation between each network element node coded by one-hot and the coordinate vector of the network element node in the high-dimensional hypergraph space. Therefore, all foreground pixel points in the complete hypergraph generated by the measurement reconstruction model can be extracted, and the gray values of the pixel points on all the layers are restored to the measured values of the network element nodes on different measurement indexes one by utilizing the mapping relation between the coordinate vectors of the pixels in the hypergraph and the network element one-hot codes, so that the restored numerical value results are assigned to the actual lack-of-measurement network elements, and the reconstruction of the random space sampling measurement of the distributed network is completed.

Compared with the prior art, the invention has the following advantages and beneficial effects:

(1) the prior art mainly performs measurement estimation through measured points around a defect point, so that the scheme is fixed, the flexibility is poor, and available information is very little. The system converts irregular topology into regular high-dimensional hypergraph according to the connection relation of network topology by vectorizing network elements. The hypergraph structure with the uniform format is convenient for subsequent data analysis and processing, random space sampling can be performed in a dynamic mode, and the problem that random space sampling in an irregular network cannot adopt a fixed model for data recovery is solved, so that the limitation of traditional static sampling is avoided.

(2) The system enables the missing point and the measured point to be fused in the hypergraph space through the hypergraph recovery generator, so that the missing point can be measured and recovered by fully utilizing the measured point with strong topological correlation in the recovery process. The parameter adjustment of the generator can be effectively driven to be optimized towards the correct direction by the authenticity discriminator.

(3) The system can combine a software defined network system and a network function virtualization technology to construct a flexible and dynamic distributed network measurement scheme, can effectively reduce the measurement data volume, improve the measurement efficiency, namely the quality, and avoid the attack of measurement points. In addition, the method can support static and dynamic space sampling and is suitable for various distributed network scenes with different types, different structures and different properties.

The following describes the implementation process of the present invention in detail with reference to specific embodiments:

the embodiment of the invention provides a measurement reconstruction method and a system for large-scale, distributed and irregular network random space sampling. As shown in fig. 1, the system includes two subsystems: a training subsystem and a reconstruction subsystem.

The training subsystem adopts historical data to establish a model for large-scale distributed network random space sampling measurement reconstruction, and comprises three functional modules: the system comprises a topological hypergraph expression module, a network measurement hypergraph module and a measurement reconstruction model training module.

And the reconstruction subsystem reconstructs a complete network measurement result by using the incomplete random measurement data in the large-scale distributed network scene.

The following describes the specific implementation methods of the above two subsystems respectively.

1. Training subsystem

The training subsystem consists of three functional modules, including: the system comprises a topological hypergraph expression module, a network measurement hypergraph module and a measurement reconstruction model training module.

For a given distributed network N, its network topology is first expressed in the form of a graph

Wherein

A set of network elements representing a distributed network N, for example: a relay node such as a router and a switch, and N represents the total number of network elements in the network.

Representing the set of all physical links between the network elements in the network N.

(1) Hypergraph expression module of topology

As shown in fig. 2, the implementation of the hypergraph expression module of the topology includes 5 steps: one-hot coding, end-to-end path sampling, K-element sub-path segmentation, node vectorization and topology hypergraphization.

Step 1: and (5) one-hot coding. As shown in fig. 3, for

Each node in the node list is assigned with a unique ID by adopting one-hot coding. Defining a binary sequence b of length N bits, each bit in the sequence corresponding to

For the coding of the ith node, let the ith bit of b be 1, i.e. b_iThe remaining bit positions are 0. Finally, N node binary codes are obtained and are expressed as { v₁,v₂,...,v_N}。

Step 2: end-to-end path sampling. For the network topology which completes one-hot coding, an end-to-end path set is obtained in a random walk mode, and the specific method is as follows:

(a) from a set of network nodes v₁,v₂,...,v_NRandomly choose one of themNode leaf v_iI.e. a network node with only one neighbor;

(b) from v_iSet of neighbor nodes

According to given strategy, selecting next hop node v_kThe node selection strategy can be pure random selection, or random selection according to the attribute distribution of the neighbor nodes, or v_iAnd the distribution of link attributes between the neighboring nodes is randomly selected, and v is_kSet of nodes to join a path { v }_i,v_k}；

(c) According to the method of (b), determining the next-hop neighbor of the last node in the set of path nodes and appending the next-hop neighbor to the set of path nodes until the selected node is a node other than v_iLeaf node v of_j；

(d) Recording path

Where, (i, j) represents the head-to-tail node number of the path, and l represents the serial number of the path with (i, j) as the head-to-tail node.

(e) Repeating (a) - (d) at least until all network nodes are traversed.

After step 2, an end-to-end path set is obtained

And step 3: and dividing the K-element sub-path. According to the end-to-end path set obtained by the method

And dividing each end-to-end path by a window with the width of K, wherein the window moving step length is s. Thus, a series of K-length sub-path sequences can be obtained, and the equal-length sub-path sequences form a K-element sub-path set

Wherein r is_mk∈{v₀,v₁,v₂,...,v_NK is {1, …, K }, M is {1, …, M }, K represents the length of the subpath, for the convenience of subsequent processing, K is an odd number, and M represents a K-element subpath set

The total number of sub-paths contained therein. The values of K and M are determined by the user according to specific needs. The setting of M requires that the set of sub-paths cover at least every node in the network.

And 4, step 4: and (6) vectorizing the nodes. As shown in fig. 4, the implementation is implemented by using a neural network, an input end of the neural network includes N neurons for receiving the N-dimensional one-hot codes of the intermediate nodes of the K-element sub-path sequence, and an output includes (K-1) N neurons corresponding to the one-hot codes of all the remaining nodes except the input node. The node vectorization model is essentially a prediction model, and one-hot codes of context nodes of the node vectorization model are predicted by using one-hot codes of intermediate nodes of the K-element sub-path sequence. Utilizing the K element sub-path set obtained in the step 3

And the back propagation algorithm may be trained to derive the predictive model. And after the model training is finished, extracting the generator as a node vectorization model. Inputting a set of nodes at the input of the model { v }₁,v₂,...,v_NOne-hot coding of each node in the model, and obtaining D-dimensional vector expression of the corresponding node from a hidden layer of the model. Thus, the set of network nodes can be rewritten as { (v)₁,u₁),(v₂,u₂),...,(v_N,u_N) In which u_iAnd representing the vector expression of the ith node, which corresponds to one-hot codes of the node one by one. The D-dimensional vector used to express the node implies structural information in the original network topology. For structurally similar, or closely related, neighboring nodes, their D-dimensional vectors will be relatively similar.

And 5: topological hypergraphization. And 4, normalizing the D-dimensional vectors of all the network nodes according to the dimension based on the vectorization result of the step 4. Based on D dimension vector set after normalization is expandedA D-dimension numerical value space is formed, and a network node set { (v)₁,u₁),(v₂,u₂),...,(v_N,u_N) Any one of the nodes v_iCorresponding to a point in the D-dimensional numerical space, the coordinates of the point are determined by the D-dimensional vector u of the node_iAnd (4) determining. Thus, the nodes in the network topology are projected to fixed positions in the D-dimensional numerical space. Further, the result obtained by projection is defined as a D-dimensional topological hypergraph, and each coordinate point in the hypergraph is regarded as a D-dimensional pixel point. These pixel points include two types: the first is the pixels projected by the network nodes, the second and the pixels not associated with any network node. The former is regarded as a foreground pixel point in the subsequent processing; the latter are considered as background pixels. Is (v)₁,u₁),(v₂,u₂),...,(v_N,u_N) Is stored in a database for use in subsequent operations on the D-dimensional vector u_iAnd reversely reducing the node into the one-hot coded node.

(2) Hypergraphization module for network measurement

After the D-dimensional topological hypergraph is obtained based on the method, the measurement data of each node in the training set is used as the gray level of the pixel point corresponding to the node in the D-dimensional topological hypergraph and is projected onto the topological hypergraph. The specific method comprises the following steps:

(a) in the network measurement, each network node measures C indexes. Thus, each measured node has a C-dimensional feature vector. And (3) carrying out normalization processing on the C-dimensional feature vectors of all the nodes one by one, and projecting the C-dimensional feature vectors to a numerical value interval of [0,1 ].

(b) And regarding the normalized C-dimensional features as the gray values of C layers on the D-dimensional topological hypergraph, namely, taking the measured data of each network node on the C-th measuring index as the gray values of the C-th layer of the corresponding pixel point of the node on the D-dimensional topological hypergraph. And setting background pixels which do not belong to any network node in the D-dimensional topological hypergraph as background gray levels, and setting transparency attributes of the background pixels as full transparency.

After the processing, the measurement data of the large-scale distributed network and the corresponding topological structure are mapped into a D-dimensional hypergraph, and the gray levels of C layers on the hypergraph correspond to C-dimensional measurement characteristic data.

(3) Measurement reconstruction model training module

The task of the measurement reconstruction model training module is to determine the parameters of the recovery model of the lacking measurement nodes. As shown in fig. 5, the implementation method includes three parts: random mask generator, hypergraph recovery generator, and true and false hypergraph discriminator.

Step 1: the random mask generator is specially used in the model training stage, and utilizes the hypergraph obtained by complete measurement to generate an incomplete measurement hypergraph with missing data points by generating a random mask for model training. The mask generation method comprises the following steps: according to the hypergraph topology obtained by complete measurement, mask pixel coordinates of a specified number W are randomly generated, and the mask pixel coordinates form a new layer, namely a mask layer. The main function of the mask layer is two-fold: eliminating data of part of measuring points in a training stage to form a lack sample; in both training and actual measurement, the mask can provide coordinate information of the measurement points and the defect points for the model. The recovery model can adjust the model parameters in a targeted manner.

Step 2: and (3) superposing the complete measurement hypergraph with the mask layer generated in the step (1), and eliminating the gray values of all layers of corresponding pixels by using the mask, thereby forming the measurement hypergraph with the missing points and the mask layer superposed.

And step 3: and simultaneously training a subsequent hypergraph recovery generator and a true and false hypergraph discriminator by using the hypergraph with the missing points and the overlay mask layer and the complete measurement hypergraph. The task of the hypergraph recovery generator is used for inputting a hypergraph with missing points and overlay mask layers and generating a predicted hypergraph of the missing parts, wherein the predicted hypergraph comprises gray values of all the layers; the authenticity hypergraph discriminator judges the authenticity of the predicted hypergraph of the missing part output from the generator using the authentic measured hypergraph.

The hypergraph recovery generator may choose to: a neural network of a codec architecture, a convolutional neural network, a fully-connected neural network, or other network architecture. In this patent, the structure adopted by the generator is not limited, and is decided by the user himself. It should be noted that no matter which neural network structure is adopted as the generator, the neural network structure only generates the recovery value of the missing point part through the mask, and does not influence or modify the data of the measured point. And superposing the hypergraph of the missing measurement part output by the generator with the input end of the hypergraph to obtain a recovered complete measurement hypergraph.

The essence of the truth hypergraph discriminator is a two-classifier, the input of which is a hypergraph with a complete size, and the output of which is two-classification judgment, namely: if the input hypergraph is a complete hypergraph obtained by real measurement, the output of the hypergraph is 0; otherwise the output is 1. As with the hypergraph restoration generator, although the hypergraph of the complete scale is input, the discriminator only discriminates the authenticity of the missing part by the mask code, ignoring the authenticity judgment of the measured point. Thus, noise interference of the measured point to the discrimination result can be avoided.

For a hypergraph training sample with missing points and an overlay mask layer, the implementation method of the hypergraph recovery generator and the missing hypergraph discriminator is as follows:

(a) initializing parameters of a hypergraph recovery generator and a lack measurement hypergraph discriminator;

(b) the hypergraph recovery generator calculates hypergraph data of the missing part according to the input data;

(c) superposing the output of the step (b) with the hypergraph with the missing point and the superposition mask layer input by the recovery generator to obtain a recovered complete measurement hypergraph;

(d) the true and false hypergraph discriminator is trained by utilizing the real complete measurement hypergraph and the recovery hypergraph obtained in the step (c), so that the true and false of the two hypergraph can be correctly distinguished;

(e) and (c) judging the authenticity of the recovered hypergraph obtained in the step (c) by using a discriminator obtained by training, and feeding the discrimination error back to the hypergraph recovery generator for parameter adjustment.

(f) Repeating (b) - (e).

2. Reconstruction subsystem

The function of the reconstruction subsystem is to recover complete network measurement data by using the result obtained by incomplete measurement in the actual measurement service. It comprises two parts: and measuring the reconstruction and the network expression of the reconstruction result.

The measurement reconstruction is composed of a hypergraph recovery generator of a measurement reconstruction model. And extracting a hypergraph recovery generator obtained by the measurement reconstruction model in the training stage for recovering incomplete measurement data. Assume that for a given network N there is an incomplete set of measurements and that the measurement points are known to be one-hot encoded with the missing points. The execution method comprises the following steps:

(a) and converting the measurement result of the measured point into the gray value of each layer on the topological hypergraph by adopting the same method as the training stage.

(b) And obtaining accurate pixel positions of the measured points and the missing points in the hypergraph according to the one-hot codes of the measured points and the missing points, and generating a mask layer according to the pixel positions, wherein the mask layer is used for marking the pixel coordinates of the missing points in the measured hypergraph. The mask layer is superimposed on top of the layer consisting of the measured properties.

(c) And (c) inputting the synthesized hypergraph obtained in the step (b) into a hypergraph recovery generator, and obtaining recovered data of the missing points, namely the gray level of each layer of the missing pixels according to generator parameters obtained by training.

(d) And replacing the gray level of the recovered missing point layer with the original input hypergraph to form a recovered complete measurement hypergraph.

And the network expression of the reconstruction result is used for restoring the recovered hypergraph format data into an actual network measurement data structure. Because the recovered data exists in a multi-layer multi-dimensional hypergraph mode, which is not suitable for the practical application requirement of network measurement data, a mode of restoring the data in the hypergraph mode into a feature expression is required, for example: by f_iRepresenting the measured characteristic value of the ith network node. The key link for restoring the hypergraph into the original characteristic expression is to reversely map pixel points in the hypergraph back to network nodes of one-hot coding. In the training stage, recording the relation between one-hot coding nodes and D-dimensional numerical value space coordinates { (v)₁,u₁),(v₂,u₂),...,(v_N,u_N)}. Therefore, the recovered complete measurement is exceededThe method for further converting into the network expression is as follows:

(a) and filtering foreground pixel points in the hypergraph, namely only keeping pixels corresponding to the network topology nodes one to one.

(b) From the D-dimensional hypergraph coordinates of the remaining pixels, according to { (v)₁,u₁),(v₂,u₂),...,(v_N,u_N) Finding the network node codes corresponding to the mapping relation one by one. For example: for pixel coordinates u in the D-dimensional numerical space_iThe corresponding network node can be found as v_i。

(c) And performing normalization on the gray values of each layer of a foreground pixel point in the hypergraph to obtain original values, and organizing the original values into nodes which are endowed with the numbers in the form of measurement vectors. For example: for node v_iDenormalization of its gray scale on each layer in the D-dimensional hypergraph is denoted as f_iAnd represents the measured attribute value of the ith node.

Therefore, finally obtaining the expression { f of the complete network measurement data₁,f₂,…,f_N}。

In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.

Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in a separate physical device or software module. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer, given the nature, function, and internal relationship of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is defined by the appended claims and their full scope of equivalents.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.

It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.

In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. The reconstruction method for the random space sampling measurement of the distributed network is characterized by comprising the following steps:

converting the network topology into a hypergraph structure;

constructing a multi-layer hypergraph according to the network topology under the hypergraph structure; the multi-layer hypergraph is used for representing a multi-dimensional measurement result of a large-scale distributed network;

2. The method for reconstructing random spatial sampling measurements for a distributed network as claimed in claim 1 wherein said converting the network topology into a hypergraph structure comprises:

respectively assigning a unique ID identifier to network element nodes in the distributed network topology and carrying out one-hot coding;

extracting an end-to-end path sequence set in the distributed network;

projecting the one-hot coded network element node to a coordinate point in a high-dimensional vector space according to the cut path segment, and expressing the network element node in the distributed network topology through a coordinate vector of the coordinate point to realize node vectorization;

converting the distributed network topology into a hypergraph formed by stretching the high-dimensional vector space according to the result of the node vectorization; wherein, each network element node in the network topology corresponds to a fixed pixel point in the hypergraph.

3. The method for reconstructing random spatial sampling measurement in distributed network according to claim 1, wherein said constructing a multilayer hypergraph according to a network topology under said hypergraph structure comprises:

constructing a plurality of layers for the network topology of the hypergraph structure; each layer corresponds to a measurement index or a measurement variable on a network element node in the distributed network;

the gray value of the pixel point on each layer reflects the measurement data of each dimension of the network element node.

4. The method for reconstructing random spatial sampling measurements in a distributed network according to claim 1, wherein training a reconstructed model according to the multi-layer hypergraph comprises:

and judging the pseudo hypergraph according to the discriminator to determine a real complete hypergraph, wherein when the discriminator cannot accurately identify the true hypergraph or the false hypergraph, the reconstruction model is trained.

5. The method for reconstructing random spatial sampling measurement in distributed network according to claim 1, wherein said recovering the incomplete measurement result in the actual measurement traffic into complete network measurement data according to the reconstruction model comprises:

6. The reconstruction system for the distributed network random space sampling measurement is characterized by comprising a training subsystem and a reconstruction subsystem, wherein the training subsystem comprises a topological hypergraph expression module, a network measurement hypergraph module and a measurement reconstruction model training module:

7. The system for reconstructing random spatial sampling measurements for a distributed network according to claim 6, wherein said topological hypergraph representation module comprises:

8. The system for reconstructing distributed network stochastic space sampling measurements according to claim 6, wherein the measurement reconstruction model training module comprises:

9. The system for reconstructing random spatial sampling measurements for a distributed network as claimed in claim 6 wherein said reconstruction subsystem comprises:

10. A computer-readable storage medium, characterized in that the storage medium stores a program, which is executed by a processor to implement the method according to any one of claims 1-5.