CN110855474B - Network feature extraction method, device, equipment and storage medium of KQI data - Google Patents

Network feature extraction method, device, equipment and storage medium of KQI data Download PDF

Info

Publication number
CN110855474B
CN110855474B CN201911000458.6A CN201911000458A CN110855474B CN 110855474 B CN110855474 B CN 110855474B CN 201911000458 A CN201911000458 A CN 201911000458A CN 110855474 B CN110855474 B CN 110855474B
Authority
CN
China
Prior art keywords
data
network
kqi
model
feature vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201911000458.6A
Other languages
Chinese (zh)
Other versions
CN110855474A (en
Inventor
杜翠凤
梁晓文
蒋仕宝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
GCI Science and Technology Co Ltd
Original Assignee
GCI Science and Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by GCI Science and Technology Co Ltd filed Critical GCI Science and Technology Co Ltd
Priority to CN201911000458.6A priority Critical patent/CN110855474B/en
Publication of CN110855474A publication Critical patent/CN110855474A/en
Application granted granted Critical
Publication of CN110855474B publication Critical patent/CN110855474B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
    • H04L41/5003—Managing SLA; Interaction between SLA and QoS
    • H04L41/5009—Determining service level performance parameters or violations of service level contracts, e.g. violations of agreed response time or mean time between failures [MTBF]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/50—Testing arrangements
    • H04L43/55—Testing of service level quality, e.g. simulating service usage

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention provides a method, a device, equipment and a storage medium for extracting network characteristics of KQI data, wherein the method is used for carrying out normalization processing on the collected KQI data of a network and sequencing the data according to a time sequence; intercepting sample data within a set time length before the current time from the sequenced KQI data; according to the sample data, obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model; the network feature extraction model comprises a CNN model used for extracting a first feature vector with a first dimension in a space dimension of sample data, an LSTM model used for extracting a second feature vector with a second dimension in a time dimension of the sample data, and an attention mechanism model pair used for fusing the first feature vector and the second feature vector; the method fully considers the spatial relation and the time dimension continuation relation of the multidimensional data of the network characteristics, comprehensively extracts the network characteristics of the KQI data, and solves the problem of low accuracy of characteristic identification.

Description

Network feature extraction method, device, equipment and storage medium of KQI data
Technical Field
The present invention relates to the field of communications technologies, and in particular, to a method, an apparatus, a device, and a storage medium for predicting user complaints of network quality.
Background
The quality of the network quality directly affects the user experience, so that the complaint data of the user can also directly reflect the quality of the network quality. In the traditional network quality optimization management, the network element index or the cell index of the speech system data of the monitoring platform is manually characterized by combining with the experience of an expert, and then the relationship between the network quality data characteristic and the complaint behavior is found out by combining with the complaint behavior of a user. The method depends heavily on expert experience, and because the speech system data of the monitoring platform has the characteristics of mass and high dimension, the traditional network quality optimization management method can not be suitable for the monitoring platform in the current era.
Aiming at the defects of the traditional network quality optimization management method, a complaint early warning model optimization method based on a random forest algorithm is proposed, and complaint early warning is realized by adopting a random forest; complaint early warning and early intervention analysis based on a big data model are also provided, so that the complaint reason can be traced. However, the above researches are all based on the network element index or the cell index of the complaint information to extract features, and the problem of hysteresis of the complaint of the user is not considered, that is, the complaint of the user is generally later than the occurrence time of the network quality problem is not considered, so that the existing method cannot extract more detailed information of the network quality, the extraction of the features is not comprehensive enough, and the accuracy of feature identification is low.
Disclosure of Invention
Based on the method, the device, the equipment and the storage medium, the network feature extraction method, the device, the equipment and the storage medium of the KQI data can fully consider the spatial relationship and the time dimension continuation relationship of the multidimensional data of the network features, comprehensively extract the network features of the KQI data, solve the problem of low feature identification accuracy, accurately find the network quality problem through the network features extracted by the method, and improve the accuracy of network quality complaint prediction.
In a first aspect, an embodiment of the present invention provides a method for extracting network features of KQI data, including:
carrying out normalization processing on the acquired KQI data of the network;
sequencing the normalized KQI data according to a time sequence;
intercepting sample data within a set time length before the current time from the sequenced KQI data;
obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model according to the sample data; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model;
extracting a first feature vector with a first dimension of the sample data through the CNN model;
extracting a second feature vector with a second dimension of the sample data through the LSTM model;
and fusing the first characteristic vector and the second characteristic vector through the attention mechanism model to obtain the network characteristics corresponding to the KQI data.
As an improvement of the above scheme, the intercepting sample data within a set time length before the current time from the sorted KQI data specifically includes:
and intercepting the network KQI data in a set time length before the current time from the sequenced KQI data by adopting a preset first sliding window.
As an improvement of the foregoing solution, the extracting, through the CNN model, the first feature vector having a first dimension of the sample data specifically includes:
setting N second sliding windows with different window lengths to traverse the sample data respectively to obtain N types of window data;
inputting N types of window data into the CNN model, acquiring the output characteristics of a first convolution layer of the CNN model as the shallow characteristics of the sample data, and acquiring the output characteristics of a second convolution layer of the CNN model as the deep characteristics of the sample data;
and fusing the shallow feature and the deep feature through a fusion layer of the CNN model to obtain the first feature vector.
As an improvement of the above scheme, the normalizing process performed on the acquired KQI data of the network specifically includes:
and normalizing each network quality index in the network KQI data according to the maximum value and the minimum value of each network quality index in the KQI data.
As an improvement of the above scheme, the network feature extraction model further includes a Flatten layer connected to the attention mechanism model and a full connection layer connected to the Flatten layer; and the fully-connected layer adopts a sigmod function as an activation function and serves as an output layer of the network feature extraction model.
As an improvement of the above, the method further comprises:
according to the sample data, obtaining a user complaint prediction result corresponding to the KQI data through a pre-established network feature extraction model; and outputting the user complaint prediction result by a full connection layer of the network feature extraction model.
As an improvement of the above solution, the method further includes the following training step of the network feature extraction model:
carrying out normalization processing on historical KQI data of a network collected in advance;
sequencing the normalized historical KQI data according to a time sequence;
traversing the sorted historical KQI data by adopting the first sliding window, and intercepting a plurality of training sample data with the set time length;
adding a prediction label to the training sample data; the prediction label is generated according to the collected user complaint data in the time length corresponding to the training sample data; the user complaint data comprises complaint codes and non-complaint codes;
and training the network feature extraction model according to the training sample data added with the prediction label.
In a second aspect, the present invention further provides a device for extracting network characteristics of KQI data, including:
the normalization processing module is used for performing normalization processing on the acquired KQI data of the network;
the sequencing module is used for sequencing the KQI data after the normalization processing according to a time sequence;
the data interception module is used for intercepting sample data within a set time length before the current moment from the sequenced KQI data;
the data extraction module is used for obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model according to the sample data; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model;
the CNN model is used for extracting a first feature vector with a first dimension of the sample data;
the LSTM model is used for extracting a second feature vector with a second dimension of the sample data;
and the attention mechanism model is used for fusing the first feature vector and the second feature vector to obtain the network features corresponding to the KQI data.
In a third aspect, the present invention further provides a network feature extraction device for KQI data, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the network feature extraction device implements the network feature extraction method for KQI data according to any one of the first aspect.
In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program, where when the computer program runs, the apparatus where the computer-readable storage medium is located is controlled to perform the network feature extraction method for KQI data according to any one of the first aspects.
Compared with the prior art, the invention has the beneficial effects that:
normalization processing is carried out on the acquired KQI data of the network; sequencing the normalized KQI data according to a time sequence; intercepting sample data within a set time length before the current time from the sequenced KQI data; according to the sample data, obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model; extracting a first feature vector with a first dimension on a spatial dimension of the sample data through the CNN model; extracting, by the LSTM model, a second feature vector having a second dimension in a time dimension of the sample data; fusing the first characteristic vector and the second characteristic vector through the attention mechanism model to obtain network characteristics corresponding to the KQI data; the method fully considers the spatial relation and the time dimension continuation relation of the multidimensional data of the network characteristics, comprehensively extracts the network characteristics of the KQI data, solves the problem of low accuracy of characteristic identification, can accurately check the network quality problem through the extracted network characteristics, and improves the accuracy of network quality complaint prediction.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a network feature extraction method for KQI data according to a first embodiment of the present invention;
FIG. 2 is a schematic diagram of data extraction provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram of a network architecture of the network feature extraction model provided by the embodiment of the present invention;
fig. 4 is a schematic diagram of a network feature extraction apparatus for KQI data according to a second embodiment of the present invention;
fig. 5 is a schematic diagram of a network feature extraction device for KQI data according to a third embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Please refer to fig. 1, which is a flowchart illustrating a network feature extraction method for KQI data according to an embodiment of the present invention; the network feature extraction method of the KQI data is executed by network feature extraction equipment of the KQI data, and specifically comprises the following steps:
s11: and normalizing the collected KQI data of the network.
In the implementation of the present invention, the KQI data refers to a key quality indicator for evaluating the quality of the network, which is aggregated by KPI data. It should be noted that KPI data can be obtained from a network manager in real time, and it is the prior art in the field to map corresponding KQI data by using KPI data, and detailed description is not provided herein. Taking a network optimization platform as an example, the KQI data comprises 460 network monitoring indexes such as a level value, a different-frequency hard switching success rate, a same-frequency hard switching success rate, a CS different-system hard switching success rate and the like; and because the data magnitude order of each index is inconsistent, the KQI data of the network is normalized in the embodiment of the invention.
Preferably, the normalizing process is performed on the acquired KQI data of the network, and specifically includes:
and normalizing each network quality index in the network KQI data according to the maximum value and the minimum value of each network quality index in the KQI data.
Taking the level value as an example, assuming that the value interval of the level value is [ -120, -50], and the current level value is-95, the normalization needs to normalize the current level value-95 in combination with the maximum value-50 and the minimum value-120 of the level value;
taking the handover success rate as an example, assuming that the value interval of the handover success rate is [ 60%, 99.9% ], and the current handover success rate is 98.8%, the normalization needs to be performed in combination with the maximum value of 99.9% and the minimum value of 60% to normalize the current handover success rate which is 98.8%.
The specific normalization formula is as follows:
Figure BDA0002241151060000061
wherein X represents the current value, XnormRepresenting the normalized value of the current value X.
S12: and sequencing the normalized KQI data according to the time sequence.
S13: and intercepting sample data within a set time length before the current time from the sequenced KQI data.
Preferably, the intercepting sample data within a set time length before the current time from the sorted KQI data specifically includes:
and intercepting the network KQI data in a set time length before the current time from the sequenced KQI data by adopting a preset first sliding window.
As shown in fig. 2, the data type of the set of KQI data is taken as the horizontal direction, and the time sequence is taken as the vertical direction, so as to obtain a 460-dimensional network monitoring index set. And then, the first sliding window is adopted to intercept the sample data in the set time length before the current time for the KQI data. In the embodiment of the present invention, the window length of the first sliding window is 2 weeks.
S14: obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model according to the sample data; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model;
extracting a first feature vector with a first dimension of the sample data through the CNN model;
extracting a second feature vector of the sample data with a second dimension through the LSTM model;
and fusing the first characteristic vector and the second characteristic vector through the attention mechanism model to obtain the network characteristics corresponding to the KQI data.
Preferably, the network feature extraction model further comprises a Flatten layer connected with the attention mechanism model and a fully connected layer connected with the Flatten layer; and the fully-connected layer adopts a sigmod function as an activation function and serves as an output layer of the network feature extraction model.
In an optional embodiment, the method further comprises:
according to the sample data, obtaining a user complaint prediction result corresponding to the KQI data through a pre-established network feature extraction model; and outputting the user complaint prediction result by a full connection layer of the network feature extraction model.
In the embodiment of the present invention, a network architecture of the network feature extraction model is shown in fig. 3, the network feature extraction model fuses a CNN model, an LSTM model, and an attention mechanism model, specifically, an input layer of the CNN model is used as a first input port of data, an input layer of the LSTM model is used as a second input port of data, the input layer of the attention mechanism model is connected to output layers of the CNN model and the LSTM model, and is configured to fuse feature vectors extracted through the CNN model and the LSTM model, and a Flatten layer connected to the attention mechanism model is configured to reduce dimensions of network features input by the attention mechanism model, and then perform user complaint prediction through a full connection layer. For example: the sample data is data with a time length of T and N + P dimensions (N + P < N). Inputting the input into a CNN model and an LST model respectively, and obtaining a 1 x (N + P) -dimensional first feature vector by N + P neurons through a series of CNN processing; after LSTM processing, the number of hidden layer units of the LSTM is N + P, and finally the hidden layer outputs a second feature vector with (N + P) x m dimensions; after the characteristic expressions of CNN and LSTM are obtained, an attention mechanism is introduced to fuse the characteristic expressions to form a fused characteristic expression f _ map, network characteristic extraction of KQI data is achieved, and prediction of user complaints can be achieved through probability operation of a full connection layer after the characteristic expressions are obtained; the network feature extraction model is preset with two complaint categories: complaint types and non-complaint types. The embodiment of the invention adopts CNN to fuse the spatial characteristics of the shallow layer and the deep layer based on the multi-dimensionality (460 dimensionality) of the network KQI data, so that the expression capability of the characteristic space is stronger and clearer, meanwhile, the time characteristic of the KQI data is extracted through the LSTM, so that the continuous relation of the data in the time dimension fully considers the space relation and the continuous relation of the time dimension of the multidimensional data of the network characteristic, the network characteristic of the KQI data is comprehensively extracted, the problem of low accuracy of characteristic identification is solved, the user complaints are predicted through the network characteristics, the relation between the user complaints and the network quality change is trained from the time dimension in consideration of the hysteresis of the user complaints, the change process perceived by the user can be more highlighted, the network quality problem can be accurately checked through the extracted network characteristics, and the network quality complaint result can be accurately predicted by adopting a network characteristic extraction model.
In an optional embodiment, the extracting, by using the CNN model, the first feature vector of the sample data having a first dimension specifically includes:
setting N second sliding windows with different window lengths to traverse the sample data respectively to obtain N types of window data;
inputting N types of window data into the CNN model, acquiring the output characteristic of a first convolution layer of the CNN model as the shallow characteristic of the sample data, and acquiring the output characteristic of a second convolution layer of the CNN model as the deep characteristic of the sample data;
and fusing the shallow feature and the deep feature through a fusion layer of the CNN model to obtain the first feature vector.
In the embodiment of the invention, convolution processing is carried out on input sample data through a CNN model, the spatial characteristics of the sample data are extracted from shallow to deep, and the characteristic extraction process of the CNN model comprises the following steps: and capturing data of different size areas in the sample data by adopting second sliding windows with different window lengths, extracting shallow features and deep features through different convolution layers, then performing dimensionality reduction operation on the shallow features and the deep features through a pooling layer of a CNN (convolutional neural network) model, and finally fusing the shallow features and the deep features through a CNN model fusion layer, namely superposing shallow feature graphs and deep feature graphs. Wherein, the fusion layer adopts an inclusion network structure. The size of the input characteristic diagram can be reduced by the fusion layer in a large hierarchy, the output channel is increased, and the accuracy of characteristic information extraction is improved from the spatial dimension. Assuming that the original input is data composed of n sample data with T as a time length, obtaining a first feature vector with 1 × n dimensions after passing through n neurons of a CNN model:
Figure BDA0002241151060000091
the feature extraction process of the LSTM model comprises the following steps:
inputting n sample data to an LSTM model, setting the number of hidden layer units of the LSTM model to be n, and finally outputting a second feature vector with dimensions of n multiplied by m by a hidden layer:
Figure BDA0002241151060000092
the feature fusion process of the attention mechanism model comprises the following steps:
after the characteristic expression of the sample data in the CNN model and the LSTM model is obtained, an attention mechanism is introduced to fuse the characteristic expression to form a fusion characteristic expression fmapI.e. the network characteristic. The specific expression process of feature fusion is represented by formulas (3), (4) and (5):
Figure BDA0002241151060000093
wherein, WαA weight matrix of m x n dimensions, bαThe bias parameters are obtained by network training and learning.
Figure BDA0002241151060000094
Is CrThe transposed item of (2). The first feature vector of the CNN model and the second feature vector of the LSTM model are integrated by formula (3) and expressed by a set of weights, and the set of weights is substantially different from each otherDifferential assignment is carried out on the input vectors of the intermediate points, the weights are normalized by adopting a softmax function, and then the normalized weights alpha are obtainediHidden layer output vector from different time points
Figure BDA0002241151060000095
Multiplying and summing to obtain the final fusion characteristic expression fmap。αiAnd fmapThe solving formula of (2) is as follows:
Figure BDA0002241151060000096
Figure BDA0002241151060000097
in the embodiment of the invention, the CNN is adopted to fuse the spatial features of the shallow layer and the deep layer, so that the spatial expression capability of the first feature vector is stronger and clearer, the loss of a large amount of local features of shallow layer data information in the CNN convolution process is avoided, and the feature extraction effect is better.
In an optional embodiment, the method further comprises the following training step of the network feature extraction model:
carrying out normalization processing on historical KQI data of a network collected in advance;
sequencing the normalized historical KQI data according to a time sequence;
traversing the sorted historical KQI data by adopting the first sliding window, and intercepting a plurality of training sample data with the set time length;
adding a prediction label to the training sample data; the prediction label is generated according to the collected user complaint data in the time length corresponding to the training sample data; the user complaint data comprises complaint codes and non-complaint codes;
and training the network feature extraction model according to the training sample data added with the prediction label.
In the implementation of the present invention, taking 2 weeks as an example of a set time length, traversing the historical KQI data according to a set step length by using a first sliding window, thereby extracting a plurality of training sample data of a two-dimensional matrix with a time length of 2 weeks, where the KQI data includes 460 indexes, and each training sample data is 336 × 460, as shown in fig. 2. The result data corresponding to the training sample data is a ratio of the number of periods predicted as QOE abnormality in a certain cell to a set time length. The normalization process of the training sample data is described above, and is not repeated here.
After a plurality of training sample data are constructed, a prediction label is added to each training sample data by taking 1 as a complaint code and 0 as a non-complaint code, model training is carried out on a pre-constructed network feature extraction model by adopting the training sample with the prediction label added, and extraction of KQI data and user complaint prediction can be realized through the trained model. Compared with the existing network characteristics for independently extracting the KQI by adopting the CNN or the LSTM, the method for extracting the KQI data by fusing the CNN and the LSTM can more completely extract the network characteristics of the KQI data, thereby improving the accuracy rate for complaint prediction.
Referring to fig. 4, a second embodiment of the present invention further provides a device for extracting network characteristics of KQI data, including:
the normalization processing module 1 is used for performing normalization processing on the acquired KQI data of the network;
the sequencing module 2 is used for sequencing the normalized KQI data according to a time sequence;
the data intercepting module 3 is used for intercepting sample data within a set time length before the current time from the sequenced KQI data;
the data extraction module 4 is configured to obtain, according to the sample data, a network feature corresponding to the KQI data through a pre-established network feature extraction model; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model;
the CNN model is used for extracting a first feature vector with a first dimension of the sample data;
the LSTM model is used for extracting a second feature vector with a second dimension of the sample data;
and the attention mechanism model is used for fusing the first feature vector and the second feature vector to obtain the network features corresponding to the KQI data.
In an alternative embodiment, the data interception module 3 comprises:
and the sliding window intercepting unit is used for intercepting the network KQI data within a set time length before the current moment from the sequenced KQI data by adopting a preset first sliding window.
In an alternative embodiment, the data extraction module 4 comprises:
the window data acquisition unit is used for setting second sliding windows with N different window lengths to respectively traverse the sample data to acquire N types of window data;
a feature obtaining unit, configured to input N types of window data into the CNN model, obtain a feature output by a first convolution layer of the CNN model, as a shallow feature of the sample data, and obtain a feature output by a second convolution layer of the CNN model, as a deep feature of the sample data;
and the feature fusion unit is used for fusing the shallow feature and the deep feature through a fusion layer of the CNN model to obtain the first feature vector.
In an optional embodiment, the normalization processing module 1 is specifically configured to perform normalization processing on each network quality indicator in the network KQI data according to a maximum value and a minimum value of each network quality indicator in the KQI data.
In an optional embodiment, the network feature extraction model further comprises a Flatten layer connected with the attention mechanism model and a full connection layer connected with the Flatten layer; and the fully-connected layer adopts a sigmod function as an activation function and serves as an output layer of the network feature extraction model.
In an alternative embodiment, the apparatus further comprises:
the complaint prediction module is used for obtaining a user complaint prediction result corresponding to the KQI data through a pre-established network feature extraction model according to the sample data; and outputting the user complaint prediction result by a full connection layer of the network feature extraction model.
In an optional embodiment, the normalization processing module 1 is further configured to perform normalization processing on historical KQI data of a network acquired in advance;
the sorting module 2 is further configured to sort the normalized historical KQI data according to a time sequence;
the data intercepting module 3 is further configured to traverse the sorted historical KQI data by using the first sliding window, and intercept a plurality of training sample data with the set time length;
the device further comprises:
the label adding module is used for adding a prediction label to the training sample data; the prediction label is generated according to the collected user complaint data in the time length corresponding to the training sample data; the user complaint data includes complaint codes and non-complaint codes;
and the model training module is used for training the network feature extraction model according to the training sample data added with the prediction label.
Fig. 5 is a schematic diagram of a network feature extraction device for KQI data according to a third embodiment of the present invention. As shown in fig. 5, the network feature extraction device of KQI data includes: at least one processor 11, such as a CPU, at least one network interface 14 or other user interface 13, a memory 15, at least one communication bus 12, the communication bus 12 being used to enable connectivity communications between these components. The user interface 13 may optionally include a USB interface, and other standard interfaces, wired interfaces. The network interface 14 may optionally include a Wi-Fi interface as well as other wireless interfaces. The memory 15 may comprise a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 15 may optionally comprise at least one memory device located remotely from the aforementioned processor 11.
In some embodiments, memory 15 stores the following elements, executable modules or data structures, or a subset thereof, or an expanded set thereof:
an operating system 151, which contains various system programs for implementing various basic services and for processing hardware-based tasks;
and (5) a procedure 152.
Specifically, the processor 11 is configured to call the program 152 stored in the memory 15, and execute the network feature extraction method for KQI data according to the foregoing embodiment, for example, step S11 shown in fig. 1. Alternatively, the processor, when executing the computer program, implements the functions of the modules/units in the above device embodiments, such as a normalization processing module.
Illustratively, the computer program may be partitioned into one or more modules/units that are stored in the memory and executed by the processor to implement the invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the network feature extraction device for the KQI data.
The network feature extraction device of the KQI data can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The network feature extraction device of the KQI data may include, but is not limited to, a processor and a memory. It will be understood by those skilled in the art that the schematic diagrams are merely examples of the network feature extraction device for the KQI data, and do not constitute a limitation of the network feature extraction device for the KQI data, and may include more or less components than those shown, or some components may be combined, or different components.
The Processor 11 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The general processor may be a microprocessor or the processor may be any conventional processor, etc., and the processor 11 is a control center of the network characteristic extraction device for the KQI data, and various parts of the network characteristic extraction device for the entire KQI data are connected by various interfaces and lines.
The memory 15 may be used to store the computer program and/or module, and the processor 11 implements various functions of the network feature extraction device of the KQI data by running or executing the computer program and/or module stored in the memory and calling the data stored in the memory. The memory 15 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, etc. Further, the memory 15 may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.
Wherein, the modules/units integrated by the network feature extraction device of the KQI data can be stored in a computer readable storage medium if they are implemented in the form of software functional units and sold or used as independent products. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments described above may be implemented. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U.S. disk, removable hard disk, magnetic diskette, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signal, telecommunications signal, and software distribution medium, etc. It should be noted that the computer readable medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable media does not include electrical carrier signals and telecommunications signals as is required by legislation and patent practice.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.

Claims (10)

1. A network feature extraction method of KQI data is characterized by comprising the following steps:
carrying out normalization processing on the acquired KQI data of the network;
sequencing the normalized KQI data according to a time sequence;
intercepting sample data within a set time length before the current time from the sequenced KQI data;
obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model according to the sample data; the network feature extraction model comprises a CNN model, an LSTM model and an attention mechanism model connected with the CNN model and the LSTM model;
extracting a first feature vector with a first dimension of the sample data through the CNN model;
extracting a second feature vector with a second dimension of the sample data through the LSTM model;
fusing the first characteristic vector and the second characteristic vector through the attention mechanism model to obtain network characteristics corresponding to the KQI data;
the final fusion signature is expressed as fmapWherein, the specific expression process of feature fusion is represented by formulas (3), (4) and (5):
Figure FDA0003459408710000011
Figure FDA0003459408710000012
Figure FDA0003459408710000013
wherein, WαIs a weight matrix of m × n dimensions, bαThe bias parameters are obtained by network training and learning,
Figure FDA0003459408710000014
is CrTransposed term of (2), CrA first feature vector is represented that represents a first feature vector,
Figure FDA0003459408710000015
representing a second feature vector;
Figure FDA0003459408710000016
fusion of a first feature vector representing the CNN model and a second feature vector representing the LSTM model, αiRepresenting the weight.
2. The method for extracting network features of KQI data according to claim 1, wherein the step of extracting sample data within a predetermined time period before the current time from the sorted KQI data comprises:
and intercepting the network KQI data in a set time length before the current time from the sequenced KQI data by adopting a preset first sliding window.
3. The method of claim 1, wherein the extracting, through the CNN model, the first feature vector of the sample data having a first dimension includes:
setting N second sliding windows with different window lengths to traverse the sample data respectively to obtain N types of window data;
inputting N types of window data into the CNN model, acquiring the output characteristics of a first convolution layer of the CNN model as the shallow characteristics of the sample data, and acquiring the output characteristics of a second convolution layer of the CNN model as the deep characteristics of the sample data;
and fusing the shallow feature and the deep feature through a fusion layer of the CNN model to obtain the first feature vector.
4. The method for extracting network features of KQI data according to claim 1, wherein the normalizing the acquired KQI data of the network specifically comprises:
and normalizing each network quality index in the network KQI data according to the maximum value and the minimum value of each network quality index in the KQI data.
5. The method for extracting network features of KQI data according to claim 2, wherein the network feature extraction model further comprises a Flatten layer connected to the attention mechanism model and a fully connected layer connected to the Flatten layer; and the fully-connected layer adopts a sigmod function as an activation function and serves as an output layer of the network feature extraction model.
6. The method for extracting network characteristics of KQI data according to claim 5, wherein the method further comprises:
according to the sample data, obtaining a user complaint prediction result corresponding to the KQI data through a pre-established network feature extraction model; and outputting the user complaint prediction result by a full connection layer of the network feature extraction model.
7. The method for extracting network features of KQI data according to claim 5, wherein the method further comprises the following steps of training the network feature extraction model:
carrying out normalization processing on historical KQI data of a network collected in advance;
sequencing the normalized historical KQI data according to a time sequence;
traversing the sorted historical KQI data by adopting the first sliding window, and intercepting a plurality of training sample data with the set time length;
adding a predictive label to the training sample data; the prediction label is generated according to the collected user complaint data in the time length corresponding to the training sample data; the user complaint data includes complaint codes and non-complaint codes;
and training the network feature extraction model according to the training sample data added with the prediction label.
8. A network feature extraction device for KQI data, comprising:
the normalization processing module is used for performing normalization processing on the acquired KQI data of the network;
the sequencing module is used for sequencing the KQI data after the normalization processing according to a time sequence;
the data interception module is used for intercepting sample data within a set time length before the current moment from the sequenced KQI data;
the data extraction module is used for obtaining network characteristics corresponding to the KQI data through a pre-established network characteristic extraction model according to the sample data; the network feature extraction model comprises a CNN model and an LSTM model, and an attention mechanism model connected with the CNN model and the LSTM model;
the CNN model is used for extracting a first feature vector with a first dimension of the sample data;
the LSTM model is used for extracting a second feature vector with a second dimension of the sample data;
the attention mechanism model is used for fusing the first feature vector and the second feature vector to obtain network features corresponding to the KQI data;
the final fusion signature is expressed as fmapWherein, the specific expression process of feature fusion is represented by formulas (3), (4) and (5):
Figure FDA0003459408710000041
Figure FDA0003459408710000042
Figure FDA0003459408710000043
wherein, WαA weight matrix of m x n dimensions, bαThe bias parameters are obtained by network training and learning,
Figure FDA0003459408710000044
is CrTransposed term of (2), CrA first feature vector is represented that represents a first feature vector,
Figure FDA0003459408710000045
representing a second feature vector;
Figure FDA0003459408710000046
fusion of a first feature vector representing the CNN model and a second feature vector representing the LSTM model, αiRepresenting the weight.
9. A network feature extraction device for KQI data, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the network feature extraction method for KQI data according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, comprising a stored computer program, wherein the computer program, when executed, controls an apparatus in which the computer-readable storage medium is located to perform the network feature extraction method for KQI data according to any one of claims 1 to 7.
CN201911000458.6A 2019-10-21 2019-10-21 Network feature extraction method, device, equipment and storage medium of KQI data Active CN110855474B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911000458.6A CN110855474B (en) 2019-10-21 2019-10-21 Network feature extraction method, device, equipment and storage medium of KQI data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911000458.6A CN110855474B (en) 2019-10-21 2019-10-21 Network feature extraction method, device, equipment and storage medium of KQI data

Publications (2)

Publication Number Publication Date
CN110855474A CN110855474A (en) 2020-02-28
CN110855474B true CN110855474B (en) 2022-06-17

Family

ID=69597013

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911000458.6A Active CN110855474B (en) 2019-10-21 2019-10-21 Network feature extraction method, device, equipment and storage medium of KQI data

Country Status (1)

Country Link
CN (1) CN110855474B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114330866B (en) * 2021-12-24 2023-11-24 江苏微皓智能科技有限公司 Data processing method, device, electronic equipment and computer readable storage medium
CN116418686A (en) * 2021-12-31 2023-07-11 华为技术有限公司 Model data processing method and device

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108062561A (en) * 2017-12-05 2018-05-22 华南理工大学 A kind of short time data stream Forecasting Methodology based on long memory network model in short-term
CN109299268A (en) * 2018-10-24 2019-02-01 河南理工大学 A kind of text emotion analysis method based on dual channel model
CN109872162A (en) * 2018-11-21 2019-06-11 阿里巴巴集团控股有限公司 A kind of air control classifying identification method and system handling customer complaint information
CN110287320A (en) * 2019-06-25 2019-09-27 北京工业大学 A kind of deep learning of combination attention mechanism is classified sentiment analysis model more

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10332056B2 (en) * 2016-03-14 2019-06-25 Futurewei Technologies, Inc. Features selection and pattern mining for KQI prediction and cause analysis

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108062561A (en) * 2017-12-05 2018-05-22 华南理工大学 A kind of short time data stream Forecasting Methodology based on long memory network model in short-term
CN109299268A (en) * 2018-10-24 2019-02-01 河南理工大学 A kind of text emotion analysis method based on dual channel model
CN109872162A (en) * 2018-11-21 2019-06-11 阿里巴巴集团控股有限公司 A kind of air control classifying identification method and system handling customer complaint information
CN110287320A (en) * 2019-06-25 2019-09-27 北京工业大学 A kind of deep learning of combination attention mechanism is classified sentiment analysis model more

Also Published As

Publication number Publication date
CN110855474A (en) 2020-02-28

Similar Documents

Publication Publication Date Title
CN110598620B (en) Deep neural network model-based recommendation method and device
CN112633962B (en) Service recommendation method and device, computer equipment and storage medium
CN112367273B (en) Flow classification method and device of deep neural network model based on knowledge distillation
CN111931809A (en) Data processing method and device, storage medium and electronic equipment
CN111310918B (en) Data processing method, device, computer equipment and storage medium
CN110855474B (en) Network feature extraction method, device, equipment and storage medium of KQI data
CN113837308A (en) Knowledge distillation-based model training method and device and electronic equipment
CN108667678A (en) A kind of O＆M Log security detection method and device based on big data
CN111199469A (en) User payment model generation method and device and electronic equipment
CN112132279A (en) Convolutional neural network model compression method, device, equipment and storage medium
CN118018260A (en) Network attack detection method, system, equipment and medium
CN114529750A (en) Image classification method, device, equipment and storage medium
CN117876018A (en) Method, device, electronic equipment and storage medium for identifying and predicting potential customers
CN108846476A (en) A kind of intelligent terminal security level classification method based on convolutional neural networks
CN112905987B (en) Account identification method, device, server and storage medium
CN115622810A (en) Business application identification system and method based on machine learning algorithm
CN115314400A (en) Network system anomaly detection method and device, electronic equipment and storage medium
CN113469406A (en) User loss prediction method combining multi-granularity window scanning and combined multi-classification
CN113723093A (en) Personnel management strategy recommendation method and device, computer equipment and storage medium
CN117058498B (en) Training method of segmentation map evaluation model, and segmentation map evaluation method and device
CN115102852B (en) Internet of things service opening method and device, electronic equipment and computer medium
CN118432952B (en) Abnormality detection method under zero trust environment, electronic equipment and storage medium
CN116708313B (en) Flow detection method, flow detection device, storage medium and electronic equipment
US11960455B2 (en) Method and apparatus for building database for retrieval, device and storage medium
CN116976418A (en) Data processing method, device, computer equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant