A method and device for identifying a network scene
Through deep learning and convolutional neural networks, the network link and position motion information are identified, combined with geographical location and application scenario models, the problem of low accuracy in network feature scene recognition in the existing technology is solved, and more efficient network optimization and users are achieved Improved experience.
Patent Information
- Application Number
- CN202311457269.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-03
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2043-11-03
AI Technical Summary
The existing network feature scenario recognition methods rely on manual qualitative judgment, with low accuracy, and cannot achieve intelligence and quantitativeization, resulting in poor network optimization results and affecting user experience and network performance.
By collecting network link status information and position motion information, combining geographical location scenario models and network application scenario models, deep learning and convolutional neural networks are used for training, network scenarios are accurately identified, and alarm information is reported in abnormal situations.
It improves the accuracy of network configuration adjustment, simplifies operation and maintenance work, reduces the inaccuracy caused by manual division, and achieves more efficient network optimization and user experience improvement.
Smart Images

Figure CN117580085B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of communication networks, and in particular, relates to a method and device for identifying network scenarios. Background Art
[0002] With the rapid development of information technology, the Internet has become an indispensable part of people's lives, affecting all aspects of people's lives and work.
[0003] In different network scenarios, network equipment configuration, user distribution, business volume and business types are all different. Therefore, in order to improve the network service quality and experience of users in the process of using the network, the network needs to identify the user's network usage scenario and configure corresponding network parameters, policies and applications for different network scenarios, so as to realize intelligent operation and maintenance based on different network scenarios and improve the intelligence of the network.
[0004] At present, researchers have conducted a lot of research on network scene recognition. At present, the real scene of the network can be identified through multi-dimensional information. However, the existing technologies are not perfect yet, and most of them have various disadvantages. First, the network feature scene recognition work is manually configured based on network planning experience. It is a coarse-grained qualitative division and cannot be used as an accurate basis for refined adjustment and optimization. Secondly, many technologies require additional information about the real space. On the one hand, they require specific hardware equipment, and on the other hand, they are difficult to popularize in every scene. In different wireless network feature scenarios, even in the face of the same network quality problem, the selection or adjustment range of parameter values involved in the network optimization process will be different, that is, different optimization models are often used in different network feature environments. Therefore, the accuracy of the judgment of the network feature scene will directly affect the final effect of network optimization and adjustment. The current scene division method cannot accurately identify network characteristics, and cannot be used as an input basis for comprehensive analysis of network quality.
[0005] The existing network feature scene recognition method is far from being intelligent and quantitative, and mainly relies on the qualitative judgment of the personal experience of network planning engineers. This manual division method cannot provide detailed guidance for optimization work, and it is even more impossible to establish a quantitative recognition model, resulting in a low accuracy rate for current network scene recognition. This will undoubtedly affect the adaptability of subsequent network optimization, network performance, and user network service quality. Summary of the invention
[0006] The purpose of the present invention is to provide a network scene recognition solution for identifying the real scene of a small local area network dedicated to a private network. The so-called private network means that the users of the network are a specific group of people, and they need to log in to identify themselves when using the network. The so-called small local area network refers to a network with a coverage radius of less than 100 meters and a number of users less than 500.
[0007] The first aspect of the present invention proposes a method for network scene identification, step S1, collecting network link status information and network location movement information of the network to be identified, using the network link status information and the network location movement information as inputs of a geographic location scene model, and the geographic location scene model outputs the geographic location scene of the network to be identified; step S2, obtaining the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, using the terminal type and the service type as inputs of a network application scene model, and the network application scene model outputs the network application scene of the network to be identified; step S3, based on the geographic location scene and the network application scene of the network to be identified, combined with preset rules, determine the final scene of the network to be identified, and report alarm information when a scene abnormality occurs.
[0008] According to the method of the first aspect of the present invention, in the step S1, the network link status information and the network position movement information of the network to be identified are collected, specifically including: using in-band network telemetry to send probe data packets that match the network status and telemetry tasks of the network to be identified, collecting multidimensional data based on network links of the network to be identified in a hop-by-hop manner, and classifying and time-correlating the multidimensional data based on network links, and storing them in a database; wherein the network links include satellite links, Ethernet links, 4G network links, 5G network links and Bluetooth communication links; the multidimensional data include link bandwidth, delay and packet loss rate; using physical sensors to collect the position, trajectory, speed, and altitude of the network scene recognition model, and time-correlating the data collected by the physical sensors and storing them in the database; wherein the physical sensors include GPS, altitude sensors and air pressure sensors.
[0009] According to the method of the first aspect of the present invention, in the step S1, the network link status information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified based on the matching probability between the network to be identified and at least one standard geographic location scenario; wherein, the training of the geographic location scenario model includes: using the PCA principal component analysis method to reduce the dimension of the network link status training data and the location movement training data to calculate the network model matrix; specifically including: performing PCA dimension reduction on the network link status training data and the location movement training data after normalization, sphericity test and chi-square test to obtain the network model matrix, and combining the time series to form a network portrait; training the geographic location scenario model by building a CNN convolutional neural network, inputting the network portrait into the CNN convolutional neural network for training to obtain a pre-trained geographic location scenario model, calculating the network portrait using the pre-trained geographic location scenario model, and performing backpropagation based on the error between the calculation result and the actual situation to fine-tune the geographic location scenario model to obtain a trained geographic location scenario model.
[0010] According to the method of the first aspect of the present invention, in the step S2, the main terminals being used in the network to be identified are selected, the main terminals refer to terminals whose number proportion in the network to be identified reaches a threshold, and the quantity information of each terminal type is stored in a database and updated regularly; the terminal traffic information of each terminal is collected, and the session information of the terminal is analyzed based on the traffic information to perceive the type of service it is using or providing; wherein the traffic information is obtained by capturing data packets transmitted in the network, including: data packet size, data packet number and data packet arrival time; the traffic information is stored in the database in time association; through deep message detection and time series analysis of the traffic information, the session information of the terminal is restored, and the type of service it is using or providing is perceived, thereby analyzing the number and proportion of the main terminals being used in the network.
[0011] According to the method of the first aspect of the present invention, in step S2, the terminal type and the service type are used as inputs of a network application scenario model, and the network application scenario model outputs the network application scenario in which the network to be identified is located; wherein the training of the network application scenario model includes: constructing a correspondence between the distribution data of the terminal type and the distribution data of the service type and the network application scenario, respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; preprocessing the sample set, normalizing the clustering matrix in the sample set by column, and inputting the clustering matrix into the network application scenario model based on deep learning for training to obtain a pre-trained network application scenario model; using the pre-trained network application scenario model to calculate the clustering matrix of the sample set, and using the error between the calculation result and the actual situation for backpropagation to fine-tune the network application scenario model to obtain the trained network application scenario model.
[0012] According to the method of the first aspect of the present invention, in step S3, the scene abnormality includes an error scene and a mutation scene: wherein the error scene refers to the geographical location being contrary to the application scene; the mutation scene refers to the network link status information and the network location collected in step S1, and the terminal type of each terminal obtained in step S2 and the service type of the service being used by each terminal undergoing data mutation.
[0013] According to the method of the first aspect of the present invention, in step S3, the recognition result format of the final scene is [geographic location, application scenario].
[0014] The second aspect of the present invention proposes a device for network scene recognition. The device includes: a first processing unit, configured to: collect network link status information and network location movement information of the network to be identified, use the network link status information and the network location movement information as inputs of the geographic location scene model, and the geographic location scene model outputs the geographic location scene of the network to be identified; a second processing unit, configured to: obtain the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, use the terminal type and the service type as inputs of the network application scene model, and the network application scene model outputs the network application scene of the network to be identified; a third processing unit, configured to: determine the final scene of the network to be identified based on the geographic location scene and the network application scene of the network to be identified, combined with preset rules, and report alarm information when a scene abnormality occurs.
[0015] According to the device of the second aspect of the present invention, the first processing unit is specifically configured to: collect the network link status information and the network position movement information of the network to be identified, specifically including: using in-band network telemetry to send probe data packets that match the network status and telemetry tasks of the network to be identified, collecting multi-dimensional data based on network links of the network to be identified in a hop-by-hop manner, and classifying and time-correlating the multi-dimensional data based on network links, and storing them in a database; wherein the network links include satellite links, Ethernet links, 4G network links, 5G network links and Bluetooth communication links; the multi-dimensional data include link bandwidth, delay and packet loss rate; using physical sensors to collect the position, trajectory, speed, and altitude of the network scene recognition model, and time-correlating the data collected by the physical sensors, and storing them in the database; wherein the physical sensors include GPS, altitude sensors and air pressure sensors.
[0016] According to the device of the second aspect of the present invention, the first processing unit is specifically configured as follows: the network link status information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified based on the matching probability between the network to be identified and at least one standard geographic location scenario; wherein the training of the geographic location scenario model includes: using the PCA principal component analysis method to reduce the dimension of the network link status training data and the location movement training data to calculate the network model matrix; specifically including: performing PCA dimension reduction on the network link status training data and the location movement training data after normalization, sphericity test and chi-square test to obtain the network model matrix, and combining the time series to form a network portrait; training the geographic location scenario model by building a CNN convolutional neural network, inputting the network portrait into the CNN convolutional neural network for training to obtain a pre-trained geographic location scenario model, calculating the network portrait using the pre-trained geographic location scenario model, and performing backpropagation based on the error between the calculation result and the actual situation to fine-tune the geographic location scenario model to obtain a trained geographic location scenario model.
[0017] According to the device of the second aspect of the present invention, the second processing unit is specifically configured to: select the main terminals being used in the network to be identified, the main terminals refer to the terminals whose number proportion in the network to be identified reaches a threshold, and store the quantity information of each terminal type in a database, and update it regularly; collect terminal traffic information of each terminal, analyze the session information of the terminal based on the traffic information, and perceive the type of service it is using or providing; wherein the traffic information is obtained by capturing data packets transmitted in the network, including: data packet size, data packet number and data packet arrival time; store the traffic information in the database in time association; restore the session information of the terminal through deep packet detection and time series analysis of the traffic information, perceive the type of service it is using or providing, and thus analyze the number and proportion of the main terminals being used in the network.
[0018] According to the device of the second aspect of the present invention, the second processing unit is specifically configured to: use the terminal type and the service type as inputs of a network application scenario model, and the network application scenario model outputs the network application scenario in which the network to be identified is located; wherein the training of the network application scenario model includes: constructing a corresponding relationship between the distribution data of the terminal type and the distribution data of the service type and the network application scenario, respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; preprocessing the sample set, normalizing the clustering matrix in the sample set by column, and inputting the clustering matrix into the network application scenario model based on deep learning for training to obtain a pre-trained network application scenario model; using the pre-trained network application scenario model to calculate the clustering matrix of the sample set, and using the error between the calculation result and the actual situation for backpropagation to fine-tune the network application scenario model to obtain the trained network application scenario model.
[0019] According to the device of the second aspect of the present invention, the third processing unit is specifically configured as follows: the situation where scene abnormalities occur includes error scenarios and mutation scenarios: wherein the error scenario refers to the geographical location being contrary to the application scenario; the mutation scenario refers to data mutations in the network link status information, the network location, the terminal type of each terminal, and the service type of the service being used by each terminal.
[0020] According to the device of the second aspect of the present invention, the recognition result format of the final scene is [geographic location, application scenario].
[0021] The third aspect of the present invention discloses an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for identifying a network scenario described in the first aspect of the present disclosure is implemented.
[0022] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying a network scenario described in the first aspect of the present disclosure is implemented.
[0023] In summary, the technical solution proposed in the present invention collects information from multiple dimensions, including network link status information, location movement information, terminal device type and service type in the network, etc., and inputs the collected information into the network scene recognition model to obtain the network scene where the device to be identified is located, so as to improve the accuracy of network configuration adjustment. This not only simplifies the workload of network operation and maintenance personnel, but also avoids the inaccuracy caused by manual division of network scenes. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0025] Figure 1 4 is a structural diagram for scene recognition according to an embodiment of the present invention.
[0026] Figure 2 FIG. 4 is a flow chart for scene recognition according to an embodiment of the present invention.
[0027] Figure 3 It is a schematic diagram for a traffic scene according to an embodiment of the present invention.
[0028] Figure 4 4 is a network status table for a traffic scenario according to an embodiment of the present invention.
[0029] Figure 5 The figure is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0031] The first aspect of the present invention provides a method for identifying a network scene. Figure 1-2 As shown, the method includes:
[0032] Step S1, collecting network link status information and network location movement information of the network to be identified, using the network link status information and the network location movement information as inputs of a geographic location scenario model, and having the geographic location scenario model output the geographic location scenario of the network to be identified.
[0033] Step S2: Obtain the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, use the terminal type and the service type as input of a network application scenario model, and have the network application scenario model output the network application scenario in which the network to be identified is located.
[0034] Step S3: Based on the geographical location scenario and the network application scenario of the network to be identified, the final scenario of the network to be identified is determined in combination with preset rules, and alarm information is reported when a scenario abnormality occurs.
[0035] Specifically, the method is used to identify the real-life scenario in which a small local area network dedicated to a private network is located. The so-called private network means that the users using the network are a specific group of people, and they need to log in to identify themselves when using the network. The so-called small local area network refers to a network with a coverage radius of less than 100 meters and a number of users of less than 500 people. The term "network scenario recognition device" will be used hereinafter to indicate a device that implements the network recognition method, and the device is deployed at the edge of the network in the form of a gateway router or similar method.
[0036] According to the method of the first aspect of the present invention, in the step S1, collecting the network link status information and the network location movement information of the network to be identified specifically includes:
[0037] In-band network telemetry is used to send probe data packets that match the network status and telemetry tasks of the network to be identified, and multi-dimensional data based on network links of the network to be identified is collected in a hop-by-hop manner. The multi-dimensional data based on network links is classified and time-correlated, and stored in a database; wherein the network links include satellite links, Ethernet links, 4G network links, 5G network links, and Bluetooth communication links.
[0038] The multidimensional data includes link bandwidth, latency and packet loss rate; physical sensors are used to collect the position, trajectory, speed and altitude of the network scene recognition model, and the data collected by the physical sensors are time-correlated and stored in the database; wherein the physical sensors include GPS, altitude sensors and air pressure sensors.
[0039] Specifically, in step S1, network link status information and network location movement information are collected, and the above information is input into a geographic location scenario model to obtain the geographic location scenario of the network. This includes:
[0040] Use in-band network telemetry and other methods to collect network link status information, including multi-dimensional information on all network links connected to the network scene recognition device. Network links include satellite links, Ethernet links, 4G, 5G, Bluetooth, etc. Multi-dimensional information includes link bandwidth, latency, packet loss rate, etc. In-band network telemetry actively sends probe data packets in an appropriate number and format that match the network status and telemetry tasks, collects relevant information in the network hop by hop, and returns this information to the network scene recognition device. This information is classified by link type and stored in a database in association with time.
[0041] Use physical sensors, including GPS, altitude sensors, air pressure sensors, etc., to analyze and collect location, trajectory, speed, altitude and other location and motion data of network scene recognition devices; store these data in a database by time association.
[0042] The network link information and location movement information are input into a preset geographic location recognition model, and the geographic location scene with the highest probability is selected based on the probability of the network to be recognized belonging to at least one geographic location scene. The geographic location scene recognition model needs to be pre-trained.
[0043] According to the method of the first aspect of the present invention, in step S1, the network link status information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified based on the matching probability between the network to be identified and at least one standard geographic location scenario.
[0044] Among them, the training of the geographic location scenario model includes: using the PCA principal component analysis method to reduce the dimension of the network link status training data and the location movement training data to calculate the network model matrix; specifically including: performing PCA dimension reduction on the network link status training data and the location movement training data after normalization, sphericity test and chi-square test to obtain the network model matrix, and combining the time series to form a network portrait; training the geographic location scenario model by building a CNN convolutional neural network, inputting the network portrait into the CNN convolutional neural network for training to obtain a pre-trained geographic location scenario model, calculating the network portrait using the pre-trained geographic location scenario model, and performing backpropagation based on the error between the calculation result and the actual situation to fine-tune the geographic location scenario model to obtain a trained geographic location scenario model.
[0045] Specifically, existing data is collected to build a training set. Relevant data in various scenarios is collected to establish the corresponding relationship between network link data, location movement data and corresponding geographical location.
[0046] The above network link status data and location movement data need to be reduced in dimension using the PCA principal component analysis method to calculate the network model matrix. After normalization, sphericity, chi-square test and other steps, the above data are PCA reduced in dimension and combined with time series to form a network portrait.
[0047] The network scene recognition model is trained by building a CNN convolutional neural network. The network image is input into the CNN convolutional neural network for training to obtain a pre-trained geographic location scene model. Further, the network image is calculated using the pre-trained geographic location scene model, and the error between the calculation result and the actual situation is back-propagated to fine-tune the geographic location scene model to obtain a trained geographic location scene model.
[0048] According to the method of the first aspect of the present invention, in the step S2, the main terminals being used in the network to be identified are selected, the main terminals refer to terminals whose number proportion in the network to be identified reaches a threshold, and the quantity information of each terminal type is stored in a database and updated regularly; the terminal traffic information of each terminal is collected, and the session information of the terminal is analyzed based on the traffic information to perceive the type of service it is using or providing; wherein the traffic information is obtained by capturing data packets transmitted in the network, including: data packet size, data packet number and data packet arrival time; the traffic information is stored in the database in time association; through deep message detection and time series analysis of the traffic information, the session information of the terminal is restored, and the type of service it is using or providing is perceived, thereby analyzing the number and proportion of the main terminals being used in the network.
[0049] Specifically, in step S2, the main terminal types connected to the network to be identified, that is, connected to the network scene identification device, and the types of services being used by the terminals are collected, and the above information is input into the network application scene model to obtain the application scene of the network. Including:
[0050] According to the device types reported by the terminal devices connected to the network to be identified, analyze the main types of terminal devices in the network. This information is reported when the terminal device is connected to the network, including the address, type, status, user identity, and total number of devices of the terminal device. Types may include: mobile phones, tablets, laptops, desktops, smart homes, etc.; status may include: standby, using services, providing services, failures, etc.; user identities may include: managers, ordinary users, etc.; based on the above data, select the main terminal devices in the network. The main terminal device types refer to the top 5 types of devices in the network; store the quantity information of each terminal device type in the database and update it regularly.
[0051] Based on the terminal traffic information collected by the network scene recognition device, the terminal session information is analyzed and restored to perceive the type of service it is using or providing. Traffic information is collected by the network device by capturing the data packets passing through the network scene recognition device, including: data packet size, quantity, arrival time, five-tuple and other information; this information is stored in the database by time association; through deep packet inspection and time series analysis of traffic information, the terminal session information is analyzed and restored to perceive the type of service it is using or providing, and the number and proportion of the main service types being used in the network are analyzed.
[0052] According to the above terminal device type information and service type information, the preset network application scenario model is called to determine the application scenario of the network scenario recognition device, such as assembly scenario, residential scenario, teaching scenario, shopping mall scenario, outdoor scenario, etc. The application scenario model needs to be trained in advance.
[0053] According to the method of the first aspect of the present invention, in step S2, the terminal type and the service type are used as inputs of a network application scenario model, and the network application scenario model outputs the network application scenario in which the network to be identified is located.
[0054] Among them, the training of the network application scenario model includes: constructing the corresponding relationship between the distribution data of the terminal type and the distribution data of the service type and the network application scenario respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; preprocessing the sample set, normalizing the clustering matrix in the sample set by column, and inputting the clustering matrix into the network application scenario model based on deep learning for training to obtain a pre-trained network application scenario model; using the pre-trained network application scenario model to calculate the clustering matrix of the sample set, and using the error between the calculation result and the actual situation for backpropagation to fine-tune the network application scenario model to obtain the trained network application scenario model.
[0055] Specifically, establish the correspondence between the above-mentioned network device type distribution and service type distribution data and network application scenarios, perform cluster analysis based on network application scenarios, form an input matrix, and construct a sample set. Preprocess the sample set and normalize the data by column. Input the input matrix into the deep learning model for training to obtain a pre-trained network application scenario model. Further, use the pre-trained network application scenario model to calculate the input matrix of the pre-processed sample set, and use the error between the calculation result and the actual situation for back propagation to fine-tune the model to obtain a trained network application scenario model.
[0056] According to the method of the first aspect of the present invention, in step S3, the scene abnormality includes an error scene and a mutation scene: wherein the error scene refers to the geographical location being contrary to the application scene; the mutation scene refers to the network link status information and the network location collected in step S1, and the terminal type of each terminal obtained in step S2 and the service type of the service being used by each terminal undergoing data mutation.
[0057] According to the method of the first aspect of the present invention, in step S3, the recognition result format of the final scene is [geographic location, application scenario].
[0058] Specifically, in step S3, the preset rules are called, and the previously identified geographic location scenarios and network application scenarios are combined to finally determine the network scenario, and a warning is issued when an abnormal scenario occurs, and the warning information is reported to the network administrator. The final scenario identification format is: [geographic location, application scenario].
[0059] Specifically, the abnormal scenario identification rule refers to: combining the geographical location scenario and network application scenario identified previously, identifying the following two abnormal scenarios.
[0060] 1. The wrong scenario appears, such as a shopping mall scenario at high altitude or a residential scenario on the water surface. This rule can be set by network operation and maintenance personnel according to actual conditions.
[0061] 2. A sudden change scenario occurs. A sudden change scenario refers to a sudden and drastic change in a certain data collected by the network scene recognition device, such as a sudden increase or decrease in a large number of devices in the network, or a sudden and drastic change in the location and motion information such as the moving speed of the network recognition device.
[0062] The second aspect of the present invention proposes a device for network scene recognition. The device includes: a first processing unit, configured to: collect network link status information and network location movement information of the network to be identified, use the network link status information and the network location movement information as inputs of the geographic location scene model, and the geographic location scene model outputs the geographic location scene of the network to be identified; a second processing unit, configured to: obtain the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, use the terminal type and the service type as inputs of the network application scene model, and the network application scene model outputs the network application scene of the network to be identified; a third processing unit, configured to: determine the final scene of the network to be identified based on the geographic location scene and the network application scene of the network to be identified, combined with preset rules, and report alarm information when a scene abnormality occurs.
[0063] According to the device of the second aspect of the present invention, the first processing unit is specifically configured to: collect the network link status information and the network position movement information of the network to be identified, specifically including: using in-band network telemetry to send probe data packets that match the network status and telemetry tasks of the network to be identified, collecting multi-dimensional data based on network links of the network to be identified in a hop-by-hop manner, and classifying and time-correlating the multi-dimensional data based on network links, and storing them in a database; wherein the network links include satellite links, Ethernet links, 4G network links, 5G network links and Bluetooth communication links; the multi-dimensional data include link bandwidth, delay and packet loss rate; using physical sensors to collect the position, trajectory, speed, and altitude of the network scene recognition model, and time-correlating the data collected by the physical sensors, and storing them in the database; wherein the physical sensors include GPS, altitude sensors and air pressure sensors.
[0064] According to the device of the second aspect of the present invention, the first processing unit is specifically configured as follows: the network link status information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified based on the matching probability between the network to be identified and at least one standard geographic location scenario; wherein the training of the geographic location scenario model includes: using the PCA principal component analysis method to reduce the dimension of the network link status training data and the location movement training data to calculate the network model matrix; specifically including: performing PCA dimension reduction on the network link status training data and the location movement training data after normalization, sphericity test and chi-square test to obtain the network model matrix, and combining the time series to form a network portrait; training the geographic location scenario model by building a CNN convolutional neural network, inputting the network portrait into the CNN convolutional neural network for training to obtain a pre-trained geographic location scenario model, calculating the network portrait using the pre-trained geographic location scenario model, and performing backpropagation based on the error between the calculation result and the actual situation to fine-tune the geographic location scenario model to obtain a trained geographic location scenario model.
[0065] According to the device of the second aspect of the present invention, the second processing unit is specifically configured to: select the main terminals being used in the network to be identified, the main terminals refer to the terminals whose number proportion in the network to be identified reaches a threshold, and store the quantity information of each terminal type in a database, and update it regularly; collect terminal traffic information of each terminal, analyze the session information of the terminal based on the traffic information, and perceive the type of service it is using or providing; wherein the traffic information is obtained by capturing data packets transmitted in the network, including: data packet size, data packet number and data packet arrival time; store the traffic information in the database in time association; restore the session information of the terminal through deep packet detection and time series analysis of the traffic information, perceive the type of service it is using or providing, and thus analyze the number and proportion of the main terminals being used in the network.
[0066] According to the device of the second aspect of the present invention, the second processing unit is specifically configured to: use the terminal type and the service type as inputs of a network application scenario model, and the network application scenario model outputs the network application scenario in which the network to be identified is located; wherein the training of the network application scenario model includes: constructing a corresponding relationship between the distribution data of the terminal type and the distribution data of the service type and the network application scenario, respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; preprocessing the sample set, normalizing the clustering matrix in the sample set by column, and inputting the clustering matrix into the network application scenario model based on deep learning for training to obtain a pre-trained network application scenario model; using the pre-trained network application scenario model to calculate the clustering matrix of the sample set, and using the error between the calculation result and the actual situation for backpropagation to fine-tune the network application scenario model to obtain the trained network application scenario model.
[0067] According to the device of the second aspect of the present invention, the third processing unit is specifically configured as follows: the situation where scene abnormalities occur includes error scenarios and mutation scenarios: wherein the error scenario refers to the geographical location being contrary to the application scenario; the mutation scenario refers to data mutations in the network link status information, the network location, the terminal type of each terminal, and the service type of the service being used by each terminal.
[0068] According to the device of the second aspect of the present invention, the recognition result format of the final scene is [geographic location, application scenario].
[0069] Specifically, the network scene recognition device of the present invention is generally deployed at the edge of the network, and the device includes a gateway router and a database system. The gateway router is used to collect the data required for recognition and perform calculations on the data. The database system is used to store the collected data.
[0070] In an optional / alternative embodiment, the network scene recognition device includes: a network link status perception unit, used to collect network link status information; a physical sensor module, used to collect device location movement information; a terminal device perception unit, used to collect terminal device type and service traffic information; a computing and reasoning unit, used to perform scene recognition; and a data storage unit, used to store the collected data.
[0071] The network link state sensing unit is used to collect network state information, and includes one or more network state interfaces between the device and the network link. The device senses the link state through the interface and collects link information, and the link information corresponds to the network link information described in step 1. The network state sensing unit will perform preliminary preprocessing on the collected raw data, organize it into a specific data format, and store it in the data storage unit through the data storage interface.
[0072] The physical sensor unit is used to collect information such as the location, trajectory, speed, altitude, etc. of the network device, corresponding to the location motion information in step 1. The physical sensor unit will perform preliminary preprocessing on the collected raw data, organize it into a specific data format, and store it in the data storage unit through the data storage interface.
[0073] The terminal device perception unit is used to perceive the terminal device type and service flow data, and restore the session information. It includes a static information interface, located between the terminal and the device, used to collect the static information of the terminal, corresponding to the terminal type information described in step 2; a dynamic information interface, located between the terminal and the device, used to collect the dynamic flow information of the terminal, corresponding to the flow information described in step 2. The service perception unit performs preliminary preprocessing on the collected data, organizes it into a specific data format, and stores it in the data storage unit through the data storage interface.
[0074] The computing and reasoning unit is used to infer and identify network scenarios, and includes a data extraction interface for extracting relevant data from the data storage unit. The computing and reasoning unit inputs the collected network link status information, location movement information, terminal type, service type and other information into the pre-trained network status model according to the above method, and infers the network scenario for reference in subsequent network operation and maintenance.
[0075] The data storage unit is used to store the collected data, and includes a data acquisition interface and a data provision interface. The data storage unit can be a distributed database used to store relevant data in the entire network.
[0076] Specific application examples
[0077] like Figure 3 As shown, in a traffic scenario, the scene recognition system designed by the present invention can be deployed on transportation tools such as high-speed trains and cars and used as a vehicle-mounted mobile Wi-Fi.
[0078] Taking the high-speed rail as an example, this device collects network status information and location movement information during the movement of the high-speed rail, and analyzes and infers the geographical location scene of the high-speed rail. If the collected network status information shows that various types of networks, such as 4G, 5G, satellite signals, etc., have normal and good signals, normal latency, low packet loss rate, and slow movement speed, it means that the train is in an urban scene; if various types of network signals are normal, but the network latency is long and there is a certain packet loss rate, it means that the train is in a rural scene; if various types of network signals are sometimes good and sometimes bad, the network latency is long, and the packet loss rate is high, it means that the train may be in a mountainous scene with many caves; if various types of network signals are poor, the network latency is long, and the packet loss rate is high, it means that the train may be in a wider water bridge scene.
[0079] Furthermore, by collecting the distribution of device types and service types in the network, we can analyze and infer the network application scenario in which the train is located. For example, if there are many mobile phone and tablet users, and most of them are using video playback or other entertainment services, it can be inferred that the train is a passenger train; if there are not many device types in the network, and there are some navigation, positioning and other users, then it can be determined that the train is a freight train.
[0080] In this embodiment, Figure 4 As shown in the figure, by pre-training the network status model under traffic scenarios, the model can accurately identify various real-life scenarios that may be encountered under traffic scenarios. In implementation, data is input into the model in real time to identify the corresponding traffic scenarios.
[0081] The third aspect of the present invention discloses an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for identifying a network scenario described in the first aspect of the present disclosure is implemented.
[0082] Figure 5 is a structural diagram of an electronic device according to an embodiment of the present invention, such as Figure 5As shown, the electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, near field communication (NFC) or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the housing of the electronic device, or an external keyboard, touch pad or mouse, etc.
[0083] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the technical solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0084] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying a network scenario described in the first aspect of the present disclosure is implemented.
[0085] In summary, the technical solution proposed in the present invention can be deployed at the edge of the network to collect network status information and terminal traffic information, so as to identify network scenarios simply, efficiently and accurately. There is no need to manually configure and adjust the network scenario recognition rules, which is more intelligent and accurate. A variety of precise scenarios can be identified, such as outdoor scenarios can be subdivided into mountainous areas, water areas, etc. The recognition process is simple and accurate, and the usage data is easily obtained within the network, without the need to deploy other equipment.
[0086] The technical solution proposed in the present invention performs perception analysis on both sides of the network equipment, and comprehensively considers the geographical location scenario and the network application scenario, so that the result is more accurate and in line with reality. The network scenario is analyzed through the network equipment type distribution, network service type distribution, network link status and other multi-dimensional data. Compared with previous technologies, it is more comprehensive and analyzed from multiple angles, reducing misjudgments caused by special circumstances.
[0087] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above-mentioned embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.
Claims
1. A method for identifying a network scene, characterized in that: The method comprises: Step S1, collecting network link status information and network location movement information of the network to be identified, using the network link status information and the network location movement information as inputs of a geographic location scenario model, and having the geographic location scenario model output a geographic location scenario of the network to be identified; Step S2: obtaining the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, taking the terminal type and the service type as inputs of a network application scenario model, and having the network application scenario model output the network application scenario in which the network to be identified is located; Step S3: based on the geographical location scenario and the network application scenario of the network to be identified, the final scenario of the network to be identified is determined in combination with preset rules, and alarm information is reported when a scenario abnormality occurs; Wherein, in the step S1, the network link state information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified according to the matching probability between the network to be identified and at least one standard geographic location scenario; wherein the training of the geographic location scenario model includes: The network link state training data and the position movement training data are subjected to dimensionality reduction by using the PCA principal component analysis method to calculate the network model matrix; specifically, the network link state training data and the position movement training data are subjected to normalization, sphericity test and chi-square test, and then subjected to PCA dimensionality reduction to obtain the network model matrix, and a network portrait is formed by combining the time series; The geographic location scene model is trained by building a CNN convolutional neural network, the network image is input into the CNN convolutional neural network for training, and a pre-trained geographic location scene model is obtained. The network image is calculated using the pre-trained geographic location scene model, and back-propagation is performed according to the error between the calculation result and the actual situation to fine-tune the geographic location scene model, so as to obtain a trained geographic location scene model; Wherein, in the step S2, the terminal type and the service type are used as inputs of a network application scenario model, and the network application scenario model outputs the network application scenario in which the network to be identified is located; wherein the training of the network application scenario model includes: Constructing corresponding relationships between the distribution data of the terminal type and the distribution data of the service type and the network application scenario, respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; Preprocessing the sample set, normalizing the clustering matrix in the sample set by column, inputting the clustering matrix into the network application scenario model based on deep learning for training, and obtaining a pre-trained network application scenario model; The clustering matrix of the sample set is calculated using the pre-trained network application scenario model, and the error between the calculated result and the actual situation is used for back propagation to fine-tune the network application scenario model to obtain the trained network application scenario model.
2. A method for identifying a network scene according to claim 1, characterized in that: In the step S1, collecting the network link status information and the network location movement information of the network to be identified specifically includes: Using in-band network telemetry to send probe data packets that match the network status and telemetry tasks of the network to be identified, collecting multi-dimensional data based on network links of the network to be identified in a hop-by-hop manner, and classifying and time-correlating the multi-dimensional data based on network links, and storing them in a database; The network links include satellite links, Ethernet links, 4G network links, 5G network links and Bluetooth communication links; the multi-dimensional data include link bandwidth, delay and packet loss rate; Using physical sensors to collect the location, trajectory, speed, and altitude of the network scene recognition model, and performing time correlation on the data collected by the physical sensors and storing them in the database; The physical sensors include GPS, altitude sensor and air pressure sensor.
3. A method for identifying a network scene according to claim 2, characterized in that: In step S2: Selecting the main terminals being used in the network to be identified, where the main terminals refer to the terminals whose number proportion in the network to be identified reaches a threshold, and storing the number information of each terminal type in a database, and updating it regularly; Collect terminal traffic information of each terminal, analyze session information of the terminal based on the traffic information, and perceive the type of service being used or provided by the terminal; The flow information is obtained by capturing data packets transmitted in the network, including: data packet size, data packet number and data packet arrival time; the flow information is stored in the database in time association; through deep packet detection and time series analysis of the flow information, the session information of the terminal is restored, and the type of service it is using or providing is perceived, thereby analyzing the number and proportion of major terminals being used in the network.
4. A method for identifying a network scene according to claim 3, characterized in that: In step S3, the abnormal scenarios include error scenarios and mutation scenarios: wherein the error scenario refers to the geographical location being contrary to the application scenario; the mutation scenario refers to data mutations in the network link status information and the network location collected in step S1, and the terminal types of the various terminals obtained in step S2 and the service types of the services being used by the various terminals.
5. A method for identifying a network scene according to claim 4, characterized in that: In step S3, the recognition result format of the final scene is [geographic location, application scenario].
6. A network scene recognition device, characterized in that: The device comprises: The first processing unit is configured to: collect network link status information and network location movement information of the network to be identified, use the network link status information and the network location movement information as inputs of a geographic location scenario model, and output the geographic location scenario of the network to be identified by the geographic location scenario model; The network link state information and the network location movement information are input into the geographic location scenario model, and the geographic location scenario model determines the geographic location scenario of the network to be identified according to the matching probability between the network to be identified and at least one standard geographic location scenario; wherein the training of the geographic location scenario model includes: The network link state training data and the position movement training data are subjected to dimensionality reduction by using the PCA principal component analysis method to calculate the network model matrix; specifically, the network link state training data and the position movement training data are subjected to normalization, sphericity test and chi-square test, and then subjected to PCA dimensionality reduction to obtain the network model matrix, and a network portrait is formed by combining the time series; The geographic location scene model is trained by building a CNN convolutional neural network, the network image is input into the CNN convolutional neural network for training, and a pre-trained geographic location scene model is obtained. The network image is calculated using the pre-trained geographic location scene model, and back-propagation is performed according to the error between the calculation result and the actual situation to fine-tune the geographic location scene model, so as to obtain a trained geographic location scene model; The second processing unit is configured to: obtain the terminal type of each terminal included in the network to be identified and the service type of the service being used by each terminal, use the terminal type and the service type as input of a network application scenario model, and output the network application scenario in which the network to be identified is located by the network application scenario model; The terminal type and the service type are used as inputs of a network application scenario model, and the network application scenario model outputs a network application scenario in which the network to be identified is located; wherein the training of the network application scenario model includes: Constructing corresponding relationships between the distribution data of the terminal type and the distribution data of the service type and the network application scenario, respectively, and performing cluster analysis on the distribution data of the terminal type and the distribution data of the service type based on the network application scenario to form a clustering matrix, thereby constructing a sample set based on the clustering matrix; Preprocessing the sample set, normalizing the clustering matrix in the sample set by column, inputting the clustering matrix into the network application scenario model based on deep learning for training, and obtaining a pre-trained network application scenario model; Calculating the clustering matrix of the sample set using the pre-trained network application scenario model, and performing back propagation using the error between the calculation result and the actual situation to fine-tune the network application scenario model to obtain the trained network application scenario model; The third processing unit is configured to: determine the final scenario of the network to be identified based on the geographical location scenario and the network application scenario of the network to be identified in combination with preset rules, and report alarm information when a scenario abnormality occurs.
7. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method for identifying a network scene according to any one of claims 1 to 5 is implemented.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying a network scene according to any one of claims 1 to 5 is implemented.
