CN109104731A

CN109104731A - Construction method, device and the computer equipment of cell scenario category classification model

Info

Publication number: CN109104731A
Application number: CN201810721971.3A
Authority: CN
Inventors: 李清亮; 唐艺龙; 莫景画; 刘津羽
Original assignee: Guangdong Haige Icreate Technology Co ltd
Current assignee: Guangdong Haige Icreate Technology Co ltd
Priority date: 2018-07-04
Filing date: 2018-07-04
Publication date: 2018-12-28
Anticipated expiration: 2038-07-04
Also published as: CN109104731B

Abstract

This application involves construction method, device, computer equipment and the storage mediums of a kind of cell scenario category classification model.This method includes: that each field data in the KPI Key Performance Indicator data to sample cell carries out bivariate correlation analysis with communication behavior field respectively, removes the field data with communication behavior field correlation lower than threshold value and obtains communication behavior characteristic；Principal component analysis is carried out to each communication behavior characteristic, obtains communication behavior characteristic value；Communication behavior waveform diagram is obtained using each communication behavior characteristic value and binding time granularity, and therefrom screens communication behavior waveform sample figure；Deep learning network model is trained using each communication behavior waveform sample figure and its corresponding scene type, the deep learning network model after training is determined as cell scenario category classification model.It can be improved the efficiency of cell scenario classification synchronized update using this method, for customization planning construction scheme, determine that Network Optimization Strategy provides important foundation.

Description

Construction method, device and the computer equipment of cell scenario category classification model

Technical field

This application involves field of communication technology, more particularly to a kind of cell scenario category classification model construction method, Device, computer equipment and storage medium.

Background technique

With the continuous development of mobile communications network, to different community carry out scene partitioning be customization planning construction scheme, Determine the important foundation of Network Optimization Strategy.

Cell scenario classification be divided in plot planning first stage of construction it has been determined that traditional mobile communications network cell The division of scene type relies primarily on network planning engineer according to the geographical environment of cell, covering factor and business characteristic etc. It is artificially divided, but the geographical environment of a cell, network condition and service feature are all continuous in subsequent use Variation, cell scenario classification is also required to synchronized update, basis is provided to adjust, optimizing network, but traditional cell scenario The partitioning technology of classification is difficult to quickly repartition cell scenario classification to adapt to the geographical environment of cell, covering factor And the variation of the factors such as business characteristic.

Summary of the invention

Based on this, it is necessary to be difficult to quickly for the partitioning technology of upper traditional cell scenario classification to cell scenario classification It carries out repartitioning the technical problem for realizing that the update of cell scenario classification is synchronous, a kind of cell scenario category classification model is provided Construction method, device, computer equipment and storage medium.

A kind of construction method of cell scenario category classification model, comprising the following steps:

Obtain sample cell KPI Key Performance Indicator data, in the KPI Key Performance Indicator data each field data with Communication behavior field carries out bivariate correlation analysis respectively, removes the Field Count for being lower than threshold value with communication behavior field correlation According to acquisition communication behavior characteristic；

Principal component analysis is carried out to each communication behavior characteristic, obtains communication behavior characteristic value；

Communication behavior waveform diagram is obtained using each communication behavior characteristic value and binding time granularity；

Communication behavior waveform sample figure is screened from each communication behavior waveform diagram, and obtains each communication behavior wave The corresponding scene type of shape sample graph；

Deep learning network model is carried out using each communication behavior waveform sample figure and corresponding scene type Training, is determined as cell scenario category classification model for the deep learning network model after training.

A kind of construction device of cell scenario category classification model, comprising:

Communication behavior characteristic obtains module, for obtaining the KPI Key Performance Indicator data of sample cell, to the pass Each field data and communication behavior field in key performance indicators data carry out bivariate correlation analysis respectively, remove and communicate Behavior field correlation obtains communication behavior characteristic lower than the field data of threshold value；

Communication behavior characteristic value acquisition module is obtained for carrying out principal component analysis to each communication behavior characteristic Obtain communication behavior characteristic value；

Communication behavior waveform diagram obtains module, for being obtained using each communication behavior characteristic value and binding time granularity Communication behavior waveform diagram；

Screening sample module for screening communication behavior waveform sample figure from each communication behavior waveform diagram, and obtains Take the corresponding scene type of each communication behavior waveform sample figure；

Partitioning model obtains module, for utilizing each communication behavior waveform sample figure and corresponding scene type pair Deep learning network model is trained, and the deep learning network model after training is determined as cell scenario category division mould Type.

A kind of computer equipment, including memory and processor, the memory are stored with computer program, the processing Device performs the steps of when executing the computer program

A kind of computer readable storage medium, is stored thereon with computer program, and the computer program is held by processor It is performed the steps of when row

Construction method, device, computer equipment and the storage medium of above-mentioned cell scenario category classification model, pass through acquisition The KPI Key Performance Indicator data of cell are obtained by the relevant field Data Dimensionality Reduction in KPI Key Performance Indicator data with communication behavior feature Communication behavior characteristic value is taken, and then obtains the communication behavior waveform diagram of cell, and be screened out from it training sample to deep learning Network model is trained, the subsequent KPI Key Performance Indicator data for obtaining cell, can be fast according to the KPI Key Performance Indicator data Cell is divided into different scene types by speed, improves the efficiency of cell scenario classification synchronized update, to customize planning construction Scheme determines that Network Optimization Strategy provides important foundation.

Detailed description of the invention

Fig. 1 is the applied environment figure of the construction method of cell scene type partitioning model in one embodiment of the invention；

Fig. 2 is the flow chart of the construction method of cell scene type partitioning model in one embodiment of the invention；

Fig. 3 is to obtain communication behavior using each communication behavior characteristic value and binding time granularity in one embodiment of the invention The flow chart of waveform diagram；

Fig. 4 is the stream that each communication behavior waveform sample figure and its corresponding scene type are obtained in one embodiment of the invention Cheng Tu；

Fig. 5 is that quantity is respectively divided in each communication behavior waveform diagram using clustering algorithm in one embodiment of the invention to be The flow chart of the different scenes cluster of target scene class number；

Fig. 6 is the flow chart of the construction method of cell scene type partitioning model in another embodiment of the present invention；

Fig. 7 is in one embodiment of the invention, and type is the communication waveforms figure of the cell of office building type；

Fig. 8 is the structure table of deep learning network model in one embodiment of the invention；

Fig. 9 is the structural schematic diagram of the construction device of cell scene type partitioning model in one embodiment of the invention；

Figure 10 is the internal structure chart of computer equipment in one embodiment of the invention.

Specific embodiment

It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not For limiting the application.

Referring to the application ring that Fig. 1, Fig. 1 are the construction method of cell scene type partitioning model in one embodiment of the invention Border figure.In Fig. 1, the construction method of the cell scenario category classification model can be applied to the building of cell scenario category classification model In system, which includes terminal 110 and server 120, and terminal 110 is connect by network with server 120.Terminal 110 can To be but not limited to personal computer, notebook electric energy, tablet computer etc..Server 120 can obtain sample from terminal 110 The KPI Key Performance Indicator data of cell obtain communication behavior waveform diagram, for training deep learning network model after treatment.

Referring to fig. 2, Fig. 2 is the process of the construction method of cell scene type partitioning model in one embodiment of the invention Figure, in the present embodiment, the construction method of cell scenario category classification model the following steps are included:

Step S210: the KPI Key Performance Indicator data of sample cell are obtained, to each field in KPI Key Performance Indicator data Data and communication behavior field carry out bivariate correlation analysis respectively, remove with communication behavior field correlation lower than threshold value Field data obtains communication behavior characteristic.

In this step, KPI Key Performance Indicator data (Key Performance Indicator, KPI) refer to cell base station net Pipe performance assessment criteria data, itself is with being described user communication quality parameter；Communication behavior field is KPI Key Performance Indicator data In field data relevant to communication behavior, can be the handling capacity of cell base station, be also possible to communicate the Field Counts such as percent of call completed According to.

Specifically, the crucial achievement of multiple cell base stations in the whole network can be exported using hour as dimension in network information system Achievement data is imitated into server, it, will be each in KPI Key Performance Indicator data after server obtains KPI Key Performance Indicator data Field data carries out correlation analysis with communication behavior field respectively, if certain field data and logical in KPI Key Performance Indicator data Letter behavior field correlation is lower than preset threshold, then deletes the field data from KPI Key Performance Indicator data, by crucial achievement It imitates after being deleted with communication behavior field correlation lower than the field data of preset threshold in achievement data, KPI Key Performance Indicator data In remaining field data as communication behavior characteristic.By bivariate correlation analysis, using communication behavior field as core The heart carries out data clusters, weeds out the noise field unrelated with communication behavior field, effectively reduces the interference of noise field, simultaneously Data volume is effectively reduced and improves computation rate.

Step S220: principal component analysis is carried out to each communication behavior characteristic, obtains communication behavior characteristic value.

It include multiple field datas relevant to communication behavior in communication behavior characteristic in this step；Server with The communication behavior characteristic of multiple cell base stations is that the progress principal component analysis of benchmark data is dropped in the whole network acquired Dimension, will obtain an one-dimensional communication behavior characteristic value after the communication behavior characteristic dimensionality reduction of each various dimensions.

Step S230: communication behavior waveform diagram is obtained using each communication behavior characteristic value and binding time granularity.

In this step, the communication behavior characteristic value binding time dimension that server will obtain generates one and time correlation Communication behavior waveform diagram.Specifically, KPI Key Performance Indicator data can pass through bivariate phase using the time as derived from dimension After the analysis of closing property, principal component analysis operation, an one-dimensional communication behavior characteristic value can be used according to KPI Key Performance Indicator data The communication feature under some time of cell is characterized, server can use the communication behavior characteristic value binding time dimension, generate Cell significant communication behavior point is presented in the two-dimensional coordinate waveform diagram of one arrangement according to the time, the two-dimensional coordinate waveform diagram Cloth.

Step S240: communication behavior waveform sample figure is screened from each communication behavior waveform diagram, and obtains each communication behavior The corresponding scene type of waveform sample figure.

In this step, multiple apparent communication behaviors of communication behavior distribution characteristics are screened from each communication behavior waveform diagram Waveform diagram stamps corresponding scene type label as training sample figure, and to each communication behavior waveform sample figure.

Step S250: using each communication behavior waveform sample figure and corresponding scene type to deep learning network model It is trained, the deep learning network model after training is determined as cell scenario category classification model.

In this step, each communication behavior waveform sample figure of server by utilizing and its corresponding scene type are as training Sample is input to deep learning network model and carries out Training.Specifically, server can be according to communication behavior waveform sample This figure is the input item of deep learning network model, and deep learning network model is according to the communication behavior waveform sample figure to input Analyzing and training is carried out, output recognition result scene type corresponding with the communication behavior waveform sample figure compares, according to right The parameter of each layer network in Corrected Deep learning network model more reversed than result.

Optionally, communication behavior characteristic includes network throughput, RRC connection indicated value, up channel physical resource Block utilization rate, down channel Physical Resource Block utilization rate, uplink service information physical resource block occupancy, downlink business information object Manage at least one of resource block occupancy and downlink Successful transmissions initial transmission number of blocks.

In the construction method of above-mentioned cell scenario category classification model, by obtaining the KPI Key Performance Indicator data of cell, By obtaining communication behavior characteristic value with the relevant field Data Dimensionality Reduction of communication behavior feature in KPI Key Performance Indicator data, and then obtain The communication behavior waveform diagram of cell is taken, and is screened out from it training sample and deep learning network model is trained, it is subsequent to obtain The KPI Key Performance Indicator data of cell are taken, it can be from the corresponding current behavior waveform diagram of the KPI Key Performance Indicator data acquisition, root Quickly cell is divided into different scene types according to the passage behavior waveform diagram, improves cell scenario classification synchronized update Efficiency for customization planning construction scheme, determines that Network Optimization Strategy provides important foundation.

Referring to Fig. 3, Fig. 3 is to be obtained in one embodiment of the invention using each communication behavior characteristic value and binding time granularity The flow chart of communication behavior waveform diagram；In the present embodiment, communication is obtained using each communication behavior characteristic value and binding time granularity The step of behavior waveform diagram, comprising the following steps:

Step S231: two-dimensional array is generated using communication behavior characteristic value and binding time granularity.

In this step, KPI Key Performance Indicator data can using the time as dimension export, by bivariate correlation analysis, After principal component analysis operation, an one-dimensional communication behavior characteristic value cell can be used according to KPI Key Performance Indicator data Communication behavior characteristic value is combined its corresponding time to generate two-dimensional array by the communication behavior feature under a time, server.

Step S232: being that ordinate generates communication behavior original graph using time granularity as abscissa, communication behavior characteristic value Picture.

In this step, server by utilizing belongs to the communication behavior characteristic value binding time under the same cell, different time Dimension, generating one by abscissa, communication behavior characteristic value of the time is the two-dimensional coordinate waveform diagram of ordinate, the two-dimensional coordinate Waveform diagram, that is, communication behavior original image.

Step S233: the noise in removal communication behavior original image, by the communication behavior original image after removal noise It is determined as communication behavior waveform diagram.

In this step, the noise in communication behavior original image includes the interference such as two-dimensional coordinates, framing mask, background colour Factor；Due to presently, there are picture Core Generator for the picture of generation, usually there is the interference such as reference axis, picture frame Factor can cause influence of noise to picture recognition, therefore, by obtaining the noise remove in communication behavior original image pure The waveform picture of image.Specifically, can use the reference axis and frame in Python programming rejecting communication behavior original image Equal interference noises, obtain the waveform picture of pure image, to improve the accuracy of communication behavior waveform diagram characterization communication behavior feature, In the subsequent training cell scenario category classification model using communication behavior waveform diagram, the accurate of cell scenario category division is improved Property.

In the present embodiment, the multidimensional data of data complexity is become to visualize picture number by generating communication behavior waveform diagram According to so that the communication behavior feature of different cells is embodied, digitization, substantially reduce data complexity, solve multidimensional number According to the inconsistent problem of dimension, while being convenient for the error correction and update in cell scenario category classification model later period.

Referring to fig. 4, Fig. 4 is to obtain each communication behavior waveform sample figure and its corresponding scene in one embodiment of the invention The flow chart of classification；In the present embodiment, communication behavior waveform sample figure is screened from each communication behavior waveform diagram, and is obtained each logical The step of letter behavior waveform sample figure corresponding scene type, comprising the following steps:

Step S241: target scene class number is obtained.

In this step, target scene class number refers to the quantity of scene type.

Step S242: according to target scene class number, each communication behavior waveform diagram is respectively divided using clustering algorithm Into the different scenes cluster that quantity is target scene class number, each scene clustering center and each scene clustering center are obtained Corresponding scene type.

In this step, after determining target scene class number, clustering algorithm can use by the similar communication of feature Behavior waveform diagram is divided into a cluster, so that it is target scene class that quantity, which is respectively divided, in each communication behavior waveform diagram In the different scenes cluster of other number, to obtain corresponding scene clustering center, and each field is determined according to communication behavior feature The scene type of scape cluster.

Step S243: using with the distance value at each scene clustering center be less than pre-determined distance value communication behavior waveform diagram as Current behavior waveform sample figure.

, can be using each scene clustering center as the center of circle in this step, pre-determined distance value is that radius draws a round region, will For communication behavior waveform diagram in each round region as current behavior waveform sample figure, multiple communication behaviors are quickly screened in realization The apparent communication behavior waveform diagram of distribution characteristics saves the cost of artificial screening training sample as training sample.

Step S244: determine that current behavior waveform sample figure is corresponding according to the corresponding scene type in each scene clustering center Scene type.

In the present embodiment, according to the unique characteristics of communication behavior waveform diagram, using clustering algorithm by multiple communication behavior waves Shape figure is divided into a certain number of scene clusterings, will be close to the more apparent communication behavior of scene clustering center, graphic feature For waveform diagram as training sample, realization quickly screens the apparent communication behavior waveform diagram of multiple communication behavior distribution characteristics as instruction Practice sample, saves the cost of artificial screening training sample, the screening of training sample is more objective and accurate, improves cell scenario class The accuracy not divided.

It is target that quantity, which is respectively divided, in each communication behavior waveform diagram using clustering algorithm in one of the embodiments, In the different scenes cluster of scene type number, the corresponding scene class in each scene clustering center and each scene clustering center is obtained Other step, comprising the following steps: randomly selecting quantity is the communication behavior waveform diagram of target scene class number as each field First cluster centre of scape cluster；Remaining communication behavior waveform diagram is calculated to the distance value of each first cluster centre, each will be remained Remaining communication behavior waveform diagram is respectively divided and in the first the smallest scene clustering of cluster centre distance value, obtains each communication row The scene clustering result in different scenes cluster is divided into for waveform diagram；The of each scene clustering is calculated according to scene clustering result Two cluster centres, if each second cluster centre is equal with each first cluster centre, using each second cluster centre as each scene Cluster centre, and determine it is each burn scene clustering center for scene type.

In the present embodiment, K communication behavior waveform diagram conduct is first randomly selected from multiple communication behavior waveform diagrams at random First cluster centre, wherein K is target scene class number, is then calculated in each communication behavior waveform diagram and the first cluster The distance of the heart is referred to communication behavior waveform diagram in the scene clustering where the first nearest cluster centre.It calculates new The average value of the communication behavior waveform diagram of each scene clustering formed obtains the second cluster centre, if adjacent twice poly- Class center does not have any variation, then clusters completion.

Further, the second cluster of each scene clustering is calculated according to scene clustering result in one of the embodiments, It is further comprising the steps of after the step of center: if each second cluster centre and each first cluster centre are unequal, by each Two cluster centres jump execution and calculate remaining communication behavior waveform diagram to each first cluster centre as the first cluster centre The step of distance value.

In the present embodiment, if adjacent cluster centre twice changes, sees the second cluster centre as first and gather Class center divides classification where each communication behavior waveform diagram to different cluster centres again, iterates to calculate cluster centre, directly There is no any variation to adjacent cluster centre twice.

Referring to Fig. 5, Fig. 5 is that each communication behavior waveform diagram is respectively divided using clustering algorithm in one embodiment of the invention The flow chart clustered to the different scenes that quantity is target scene class number in the present embodiment, each will be led to using clustering algorithm Letter behavior waveform diagram is respectively divided in the different scenes cluster that quantity is target scene class number, obtains in each scene clustering The step of heart and the corresponding scene type in each scene clustering center, comprising the following steps:

Step S310: randomly selecting quantity is the communication behavior waveform diagram of target scene class number as each scene clustering The first cluster centre.

Step S320: calculating remaining communication behavior waveform diagram to the distance value of each first cluster centre, will be each remaining Communication behavior waveform diagram is respectively divided with the first the smallest scene clustering of cluster centre distance value, obtains each communication behavior wave Shape figure is divided into the scene clustering result in different scenes cluster.

Step S330: calculating the second cluster centre of each scene clustering according to scene clustering result, judges in the first cluster Whether the heart is equal with the second cluster centre, if the first cluster centre is equal with the second cluster centre, go to step S340；If First cluster centre and the second cluster centre are unequal, then go to step S350.

Step S340: using each second cluster centre as each scene clustering center, and determine each scene clustering center for Scene type.

Step S350: using the second cluster centre as the first cluster centre, go to step S320.

In the present embodiment, K communication behavior waveform diagram conduct is first randomly selected from multiple communication behavior waveform diagrams at random First cluster centre, wherein K is target scene class number, is then calculated in each communication behavior waveform diagram and the first cluster The distance of the heart is referred to communication behavior waveform diagram in the cluster where the first nearest cluster centre.Calculate new formed Each cluster communication behavior waveform diagram average value obtain the second cluster centre, if adjacent cluster centre twice does not have There is any variation, then clusters completion, it is another centered on the second cluster centre if adjacent cluster centre twice changes The secondary division for carrying out communication behavior waveform diagram scene type, to guarantee the convergence of cluster and be carried out to communication behavior waveform diagram The accuracy of clustering.

In one of the embodiments, obtain sample cell KPI Key Performance Indicator data the step of after, further include with Lower step: zero filling operation is carried out to the null value in KPI Key Performance Indicator data, and deletes the repetition in KPI Key Performance Indicator data ?.

In the present embodiment, it is complete but exist empty often to there is data field in the KPI Key Performance Indicator data of sample cell There is the problems such as identical duplicate keys therewith in value, partial data, server can be to the KPI Key Performance Indicator acquired Null value in data carries out zero filling operation, to entry deletion is repeated, utmostly to reduce the influence of noise of data.

It is the process of the construction method of cell scene type partitioning model in another embodiment of the present invention referring to Fig. 6, Fig. 6 Figure；In the present embodiment, the construction method of cell scenario category classification model, comprising the following steps:

Step S401: obtain sample cell KPI Key Performance Indicator data, to the null value in KPI Key Performance Indicator data into Row zero filling operation, and delete the duplicate keys in KPI Key Performance Indicator data.

In this step, server exports the cell base station of certain year the whole network using hour as the annual key of dimension from system Performance indicators data, wherein in each KPI Key Performance Indicator data include field include: time, cell name CGI (Cell Global Identifier, Cell Global Identification), antenna angle, bandwidth of cell, working frequency range, teledata handling capacity, Ascending physical signal resource block average throughput, downlink physical resource block average throughput, RRC connection maximum number, RRC connection foundation are asked Ask number, RRC connection be successfully established number, ascending physical signal resource block average utilization, downlink physical resource block average utilization, Uplink service information physical resource block occupancy, downlink business information physical resource block occupancy, Physical Resource Block uplink are average Interference level average value, QPSK mode downlink Successful transmissions initial transmission number of blocks, 16QAM mode downlink Successful transmissions initially pass Defeated number of blocks, 64QAM mode downlink Successful transmissions initial transmission number of blocks, E-RAB drop rate, wireless drop rate etc..It is acquiring To after great amount of samples KPI Key Performance Indicator data, server carries out the basic cleaning of data to sample KPI Key Performance Indicator data, Zero in sample KPI Key Performance Indicator data is subjected to zero filling operation, duplicate keys therein are deleted, to the full extent Reducing data noise influences.

Step S402: bivariate is carried out respectively with communication behavior field to each field data in KPI Key Performance Indicator data Correlation analysis removes the field data with communication behavior field correlation lower than threshold value and obtains communication behavior characteristic.

In this step, field data is numerous in KPI Key Performance Indicator data, and server can be to KPI Key Performance Indicator data In field data carry out data clusters, that is to say, that with the communication feature of network throughput, percent of call completed etc. and user in cell The relevant field of behavior is core cluster, cancelling noise field.Specifically, to each field data in KPI Key Performance Indicator data Bivariate correlation analysis is carried out respectively with communication behavior field, by the time in KPI Key Performance Indicator data, cell name CGI, antenna angle, bandwidth of cell, working frequency range etc. and communication behavior field correlation lower than threshold value, will not be to cell communication behavior The field data that feature generates the variation of big degree is rejected, and reduces the influence of data noise, wherein communication behavior field can wrap Include network throughput, the field datas such as wireless drop rate and RRC successful connection number.

Step S403: principal component analysis is carried out to each communication behavior characteristic, obtains communication behavior characteristic value.

In this step, KPI Key Performance Indicator data removal with after the lower field data of communication behavior field correlation, most 11, field data conduct research of the correlation between any two greater than 0.5 threshold value, this 11 field datas are filtered out eventually Including request number of times is established in network throughput, RRC connection maximum number, RRC connection, RRC connection is successfully established number, up channel Physical Resource Block utilization rate, down channel Physical Resource Block utilization rate, uplink service information physical resource block occupancy, lower industry Business information physical resource block occupancy, QPSK mode downlink Successful transmissions initial transmission number of blocks, 16QAM mode downlink successfully pass Defeated initial transmission number of blocks and 64QAM mode downlink Successful transmissions initial transmission number of blocks.By the cell base station of certain year the whole network It is the communication behavior characteristic in the KPI Key Performance Indicator data of dimension as reference data using hour, carries out principal component point Analysis obtains communication behavior characteristic value.

Step S404: two-dimensional array is generated using communication behavior characteristic value and binding time granularity.

In this step, KPI Key Performance Indicator data can divide using hour as derived from dimension by bivariate correlation After analysis, principal component analysis operation, one-dimensional communication behavior characteristic value cell can be used according to KPI Key Performance Indicator data Communication feature under a time, server will belong to communication behavior characteristic value knot of the same cell on the same day under different time Its corresponding time generation two-dimensional array is closed, generates two-dimensional array according to sequence arrangement in 24 hours.

Step S405: being that ordinate generates communication behavior original graph using time granularity as abscissa, communication behavior characteristic value Picture.

In this step, it is vertical that server, which generates one by abscissa, communication behavior characteristic value of the time according to two-dimensional array, The two-dimensional coordinate waveform diagram of the 24*24 of coordinate, the two-dimensional coordinate waveform diagram, that is, communication behavior original image, and then obtain certain year The communication behavior original image of the characterization communication behavior feature of every day of all cells of the whole network.

Step S406: the noise in removal communication behavior original image, by the communication behavior original image after removal noise It is determined as communication behavior waveform diagram.

Step S407: target scene class number is obtained.

In this step, traditional scene type number is usually in 6 classes to 13 classes, on this basis, can use cluster and calculates Each communication behavior waveform diagram is divided into the scene type of multiclass, obtains respectively with scene type number for 6 classes to 13 classes by method 6 classes to 13 classes cluster result, compare the communication behavior waveform diagram in each cluster result in each scene type closeness, The indexs such as the difference between in all kinds of scene types, the most by the best scene type number classification of the overall quality of cluster result Target scene class number.Finally, the best scene type number classification of the overall quality of cluster result is 9 classes, by target field Scape class number is determined as 9 classes.

Step S408: according to target scene class number, each communication behavior waveform diagram is respectively divided using clustering algorithm Into the different scenes cluster that quantity is target scene class number, each scene clustering center and each scene clustering center are obtained Corresponding scene type.

In this step, after communication behavior waveform diagram is carried out cluster acquisition scene clustering center, for each scene clustering The communication behavior feature at center determines the scene type at each scene clustering center, and the final nine class scene types that define are respectively as follows: Undaform, failed-type, overtime work area's office building type, office building type, discrete type, market type, bar type, lodging zone type and rural area Type.Referring to Fig. 7, Fig. 7 is in one embodiment of the invention, and type is the communication waveforms figure of the cell of office building type, in figure, 0 point Communication behavior to 8 sections is less, and 8 points to 19 points of communication behavior begins to ramp up rear and the sustained periods of time time peak value, and 19 O'clock it is in sharp fall trend to 0 o'clock communication behavior, meets the communication behavior feature of office building cell.

Step S409: using with the distance value at each scene clustering center be less than pre-determined distance value communication behavior waveform diagram as Current behavior waveform sample figure.

Step S410: determine that current behavior waveform sample figure is corresponding according to the corresponding scene type in each scene clustering center Scene type.

Step S411: using each communication behavior waveform sample figure and corresponding scene type to deep learning network model It is trained, the deep learning network model after training is determined as cell scenario category classification model.

In this step, after obtaining each communication behavior waveform sample figure and its corresponding scene type, in this, as weak Supervised training sample is input to deep learning network model and carries out Training, specifically, each communication behavior waveform sample Scheme input item as deep learning network model, the corresponding scene type of each communication behavior waveform sample figure is as depth The output item of network model is practised, deep learning network model carries out analyzing and training to the communication behavior waveform sample figure of input, defeated Recognition result scene type comparison corresponding with communication behavior waveform sample figure out, learns according to the reversed Corrected Depth of comparing result The parameter of each layer network in network model.Wherein, the structure of deep learning network model is as shown in figure 8, the deep learning network Using maximum pond layer and two kinds of random pool layer, the step-length of random pool layer is set as pond layer in modelDepth The number of iterations for practising the training of network model is set as 100000 times, and optimizer uses Adadelta optimizer, learning rate It is set as 0.001, step-length is set as 0.0005.By using the mode of random pool layer, increase the deep learning network model Fuzzy diagnosis effect, improve the robustness of deep learning network model.

In the present embodiment, by obtain cell KPI Key Performance Indicator data, by KPI Key Performance Indicator data with communication The relevant field Data Dimensionality Reduction of behavioural characteristic obtains communication behavior characteristic value, and then filters out communication behavior waveform sample figure to depth Degree learning network model is trained, the subsequent KPI Key Performance Indicator data for obtaining cell, can be according to the KPI Key Performance Indicator Cell is quickly divided into different scene types by data, improves the efficiency of cell scenario classification synchronized update, for customization rule It draws construction scheme, determine that Network Optimization Strategy provides important foundation.

The deep learning network model after training is determined as cell scenario category division mould in one of the embodiments, It is further comprising the steps of after the step of type: to obtain the first KPI Key Performance Indicator data of cell to be divided, closed from described first The first communication behavior characteristic of the cell to be divided is obtained in key performance indicators data；Utilize default principal component coefficient square Battle array carries out principal component dimensionality reduction to each first communication behavior related data, obtains the first communication behavior of the cell to be divided Characteristic value；The first communication behavior waveform diagram is generated using each first communication behavior characteristic value and binding time granularity；According to The first communication behavior waveform diagram obtains the scene type of the cell to be divided.

In the present embodiment, default principal component coefficient matrix can be led in the communication behavior characteristic to sample cell When constituent analysis, one-dimensional element is filtered out according to the coefficient of principal component analysis and is obtained；Server obtains in certain period of time wait draw The KPI Key Performance Indicator data for dividing cell extract field data conduct relevant to communication behavior from KPI Key Performance Indicator data Communication behavior characteristic, and by the first KPI Key Performance Indicator of cell data to be divided the first communication behavior characteristic with Default principal component coefficient matrix is multiplied, and the first communication behavior characteristic dimensionality reduction is obtained corresponding the first one-dimensional communication behavior Characteristic value is obtaining cell to be divided by combining the corresponding time to generate communication behavior waveform diagram communication behavior characteristic value Communication behavior waveform diagram after, the graphic feature of communication behavior waveform diagram is input to the first communication behavior of cell to be divided Waveform diagram is input to preparatory trained deep learning network model, obtains the scene type of cell to be divided.By to depth The utilization of learning network model, can be in the case where getting the KPI Key Performance Indicator data of cell, quickly to the field of cell Scape type is divided, real when the geographical environment of cell, network condition and service feature change in subsequent use The update of existing cell scenario classification is synchronous, and then optimizes and build for subzone network and provide decision assistant.Meanwhile passing through acquisition The KPI Key Performance Indicator data of cell are obtained by the relevant field Data Dimensionality Reduction in KPI Key Performance Indicator data with communication behavior feature Communication behavior characteristic value is obtained, and then obtains the communication behavior waveform diagram of cell according to communication behavior characteristic value, with communication behavior wave Partitioning standards of the shape figure as the scene type of cell, so that the accuracy of the division of cell scenario classification greatly improves, it is fixed Planning construction scheme processed determines that Network Optimization Strategy provides important foundation.

In one of the embodiments, in the scene class for the cell to be divided for obtaining the output of cell scenario category classification model After other step, the network state parameters of cell to be divided can also be adjusted, according to the scene type of cell to be divided with excellent Change the outfit of cell base station, the network state parameters that the scene type of cell to be divided adjusts cell to be divided specifically include following Step: corresponding FEATURE service collection is determined according to the scene type of Target cell；It obtains FEATURE service and concentrates each FEATURE service The mass parameter and a reference value of the service-aware factor；According to the mass parameter of the service-aware factor and a reference value from business sense Know and determine perception superiority and inferiority impact factor in the factor, and matches the target network parameter for influencing perception superiority and inferiority impact factor；According to mesh The preset estimated performance index for marking cell adjusts target network parameter.

In the present embodiment, by monitor in real time daily each cell base station KPI Key Performance Indicator data the case where, to field The cell of scape classification variation carries out real-time scene type division, while obtaining the scene class according to the division result of scene type FEATURE service collection under not, determine influence terminal user use FEATURE service experience perception superiority and inferiority impact factor, and combine to The estimated performance index adjustment of division of cells influences the perception superiority and inferiority impact factor network parameter, realizes the scene class for being directed to cell The perception superiority and inferiority impact factor of FEATURE service under not, precisely adjusts key network parameter, is effectively improved the clothes of subzone network Business performance, improves network optimization effect, realizes fining adjusting and optimizing network.

The FEATURE service under same scene type is the same in one of the embodiments, under different scenes classification FEATURE service be different, therefore, corresponding FEATURE service collection, specific steps packet under every kind of scene type can be obtained in advance It includes: under the scene type for obtaining Target cell, the amount of access and flow of all kinds of business in full dose business；According to all kinds of business Amount of access and flow calculate separately the weighted value of all kinds of business in full dose business；It is more than the industry of preset threshold by weighted value accounting Business is determined as corresponding FEATURE service under the scene type of Target cell, obtains FEATURE service collection.

Specifically, all types of business access amounts and flow in full dose business are counted under all types of scene types, based on every The difference of class amount of access and flow obtains the weighted value of all kinds of business under all types of scene types, according to the size of weighted value, Selection business weighted value accounting is more than that the business of certain threshold value is the FEATURE service of the scene, forms FEATURE service collection.By true The FEATURE service collection of fixed each scene type can quickly determine the cell when obtaining the scene type classification results of cell FEATURE service collection improve system to obtain the perception superiority and inferiority impact factor and corresponding network parameter under the scene type The optimization efficiency of system.

It should be understood that although each step in the flow chart of Fig. 2 to Fig. 6 is successively shown according to the instruction of arrow, But these steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly state otherwise herein, these There is no stringent sequences to limit for the execution of step, these steps can execute in other order.Moreover, Fig. 2 is into Fig. 6 At least part step may include that perhaps these sub-steps of multiple stages or stage are not necessarily same to multiple sub-steps One moment executed completion, but can execute at different times, and the execution in these sub-steps or stage sequence is also not necessarily Be successively carry out, but can at least part of the sub-step or stage of other steps or other steps in turn or Alternately execute.

According to the construction method of above-mentioned cell scenario category classification model, the present invention also provides a kind of cell scenario classifications to draw The construction device of sub-model is carried out with regard to the embodiment of the construction device of cell scenario category classification model of the invention detailed below Explanation.

Referring to Fig. 9, Fig. 9 is that the structure of the construction device of cell scene type partitioning model in one embodiment of the invention is shown It is intended to, a kind of construction device of cell scenario category classification model, comprising: communication behavior characteristic obtains module 510, communication Behavioural characteristic value obtains module 520, communication behavior waveform diagram obtains module 530, screening sample module 540 and partitioning model and obtains Module 550, in which:

Communication behavior characteristic obtains module 510, for obtaining the KPI Key Performance Indicator data of sample cell, to key Each field data and communication behavior field in performance indicators data carry out bivariate correlation analysis respectively, remove and communicate row Field data for field correlation lower than threshold value obtains communication behavior characteristic；

Communication behavior characteristic value acquisition module 520 is obtained for carrying out principal component analysis to each communication behavior characteristic Communication behavior characteristic value；

Communication behavior waveform diagram obtains module 530, for being obtained using each communication behavior characteristic value and binding time granularity Communication behavior waveform diagram；

Screening sample module 540 for screening communication behavior waveform sample figure from each communication behavior waveform diagram, and obtains The corresponding scene type of each communication behavior waveform sample figure；

Partitioning model obtains module 550, for utilizing each communication behavior waveform sample figure and corresponding scene type pair Deep learning network model is trained, and the deep learning network model after training is determined as cell scenario category division mould Type.

Communication behavior waveform diagram obtains module 530 and is used for using communication behavior characteristic value simultaneously in one of the embodiments, Binding time granularity generates two-dimensional array；It is that ordinate generates communication row using time granularity as abscissa, communication behavior characteristic value For original image；The noise in communication behavior original image is removed, the communication behavior original image after removal noise is determined as Communication behavior waveform diagram.

Screening sample module 540 is for obtaining target scene class number in one of the embodiments,；According to target field Scape class number, using clustering algorithm by each communication behavior waveform diagram be respectively divided quantity be target scene class number not With in scene clustering, obtaining the corresponding scene type in each scene clustering center and each scene clustering center；It will gather with each scene The distance value at class center is less than the communication behavior waveform diagram of pre-determined distance value as current behavior waveform sample figure；According to each scene The corresponding scene type of cluster centre determines the corresponding scene type of current behavior waveform sample figure.

Screening sample module 540 is target scene class number for randomly selecting quantity in one of the embodiments, First cluster centre of the communication behavior waveform diagram as each scene clustering；Remaining communication behavior waveform diagram is calculated to each first The distance value of cluster centre each remaining communication behavior waveform diagram is respectively divided the smallest with the first cluster centre distance value In scene clustering, the scene clustering result that each communication behavior waveform diagram is divided into different scenes cluster is obtained；It is poly- according to scene Class result calculates the second cluster centre of each scene clustering, will when each second cluster centre is equal with each first cluster centre Each second cluster centre as each scene clustering center, and determine each scene clustering center for scene type.

Screening sample module 540 is used in each second cluster centre and each first cluster in one of the embodiments, When the heart is unequal, using each second cluster centre as the first cluster centre, jumps execution and calculate remaining communication behavior waveform diagram To each first cluster centre distance value the step of.

Communication behavior characteristic obtains module 510 and is used for KPI Key Performance Indicator data in one of the embodiments, In null value carry out zero filling operation, and delete the duplicate keys in KPI Key Performance Indicator data.

Communication behavior characteristic includes network throughput, RRC connection indicated value, uplink in one of the embodiments, Channel Physical resource block utilization rate, down channel Physical Resource Block utilization rate, uplink service information physical resource block occupancy, under Industry business at least one of information physical resource block occupancy and downlink Successful transmissions initial transmission number of blocks.

The specific restriction of construction device about cell scenario category classification model may refer to above for cell field The restriction of the construction method of scape category classification model, details are not described herein.The building of above-mentioned cell scenario category classification model fills Modules in setting can be realized fully or partially through software, hardware and combinations thereof.Above-mentioned each module can be in the form of hardware It is embedded in or independently of the storage that in the processor in computer equipment, can also be stored in a software form in computer equipment In device, the corresponding operation of the above modules is executed in order to which processor calls.

In one embodiment, a kind of computer equipment is provided, which can be server, internal junction Composition can be as shown in Figure 10.The computer equipment include by system bus connect processor, memory, network interface and Database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory packet of the computer equipment Include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating The database of machine equipment is used to store the KPI Key Performance Indicator data of the whole network cell.The network interface of the computer equipment be used for External terminal passes through network connection communication.To realize that a kind of cell scenario classification is drawn when the computer program is executed by processor The construction method of sub-model.

It will be understood by those skilled in the art that structure shown in Figure 10, only part relevant to application scheme The block diagram of structure, does not constitute the restriction for the computer equipment being applied thereon to application scheme, and specific computer is set Standby may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.

In one of the embodiments, the present invention also provides a kind of computer equipment, including memory and processor, deposit Computer program is stored in reservoir, which performs the steps of when executing computer program

Obtain sample cell KPI Key Performance Indicator data, in KPI Key Performance Indicator data each field data with communicate Behavior field carries out bivariate correlation analysis respectively, removes the field data with communication behavior field correlation lower than threshold value and obtains Obtain communication behavior characteristic；

Communication behavior waveform sample figure is screened from each communication behavior waveform diagram, and obtains each communication behavior waveform sample figure Corresponding scene type；

Deep learning network model is trained using each communication behavior waveform sample figure and corresponding scene type, Deep learning network model after training is determined as cell scenario category classification model.

Processor executes computer program and realizes using each communication behavior characteristic value and combine in one of the embodiments, When time granularity obtains the step of communication behavior waveform diagram, implements following steps: using communication behavior characteristic value and combining Time granularity generates two-dimensional array；It is that ordinate generates communication behavior original using time granularity as abscissa, communication behavior characteristic value Beginning image；The noise in communication behavior original image is removed, the communication behavior original image after removal noise is determined as communicating Behavior waveform diagram.

Processor executes computer program and realizes to screen from each communication behavior waveform diagram and leads in one of the embodiments, Letter behavior waveform sample figure, and when obtaining the step of the corresponding scene type of each communication behavior waveform sample figure, specific implementation with Lower step: target scene class number is obtained；According to target scene class number, using clustering algorithm by each communication behavior waveform Figure is respectively divided in the different scenes cluster that quantity is target scene class number, obtains each scene clustering center and each field The corresponding scene type of scape cluster centre；The communication behavior wave of pre-determined distance value will be less than with the distance value at each scene clustering center Shape figure is as current behavior waveform sample figure；Current behavior waveform sample is determined according to the corresponding scene type in each scene clustering center The corresponding scene type of this figure.

Processor execution computer program, which is realized, in one of the embodiments, utilizes clustering algorithm by each communication behavior wave Shape figure is respectively divided in the different scenes cluster that quantity is target scene class number, obtains each scene clustering center and each When the step of the corresponding scene type in scene clustering center, implement following steps: randomly selecting quantity is target scene class First cluster centre of the communication behavior waveform diagram of other number as each scene clustering；Remaining communication behavior waveform diagram is calculated to arrive Each remaining communication behavior waveform diagram is respectively divided and the first cluster centre distance value the distance value of each first cluster centre In the smallest scene clustering, the scene clustering result that each communication behavior waveform diagram is divided into different scenes cluster is obtained；According to Scene clustering result calculates the second cluster centre of each scene clustering, if each second cluster centre and each first cluster centre phase Deng, then using each second cluster centre as each scene clustering center, and determine each scene clustering center for scene type.

Each second cluster is also performed the steps of when if processor executing computer program in one of the embodiments, Center and each first cluster centre are unequal, then using each second cluster centre as the first cluster centre, it is surplus to jump execution calculating Remaining communication behavior waveform diagram to each first cluster centre distance value the step of.

It is also performed the steps of when processor executes computer program in one of the embodiments, and Key Performance is referred to The null value marked in data carries out zero filling operation, and deletes the duplicate keys in KPI Key Performance Indicator data.

The present invention also provides a kind of computer readable storage mediums in one of the embodiments, are stored thereon with meter Calculation machine program, performs the steps of when computer program is executed by processor

Computer program is executed by processor realization using each communication behavior characteristic value and ties in one of the embodiments, When closing the step of time granularity acquisition communication behavior waveform diagram, implements following steps: using communication behavior characteristic value and tying It closes time granularity and generates two-dimensional array；It is that ordinate generates communication behavior using time granularity as abscissa, communication behavior characteristic value Original image；The noise in communication behavior original image is removed, the communication behavior original image after removal noise is determined as leading to Letter behavior waveform diagram.

Computer program is executed by processor realization and screens from each communication behavior waveform diagram in one of the embodiments, Communication behavior waveform sample figure, and when obtaining the step of the corresponding scene type of each communication behavior waveform sample figure, specific implementation Following steps: target scene class number is obtained；According to target scene class number, using clustering algorithm by each communication behavior wave Shape figure is respectively divided in the different scenes cluster that quantity is target scene class number, obtains each scene clustering center and each The corresponding scene type in scene clustering center；The communication behavior of pre-determined distance value will be less than with the distance value at each scene clustering center Waveform diagram is as current behavior waveform sample figure；Current behavior waveform is determined according to the corresponding scene type in each scene clustering center The corresponding scene type of sample graph.

Computer program is executed by processor realization using clustering algorithm for each communication behavior in one of the embodiments, Waveform diagram be respectively divided quantity be target scene class number different scenes cluster in, obtain each scene clustering center and When the step of the corresponding scene type in each scene clustering center, implement following steps: randomly selecting quantity is target scene First cluster centre of the communication behavior waveform diagram of class number as each scene clustering；Calculate remaining communication behavior waveform diagram To the distance value of each first cluster centre, each remaining communication behavior waveform diagram is respectively divided and the first cluster centre distance It is worth in the smallest scene clustering, obtains the scene clustering result that each communication behavior waveform diagram is divided into different scenes cluster；Root The second cluster centre of each scene clustering is calculated according to scene clustering result, if each second cluster centre and each first cluster centre phase Deng, then using each second cluster centre as each scene clustering center, and determine each scene clustering center for scene type.

If it is poly- also to perform the steps of each second when computer program is executed by processor in one of the embodiments, Class center and each first cluster centre are unequal, then using each second cluster centre as the first cluster centre, jump and execute calculating Remaining communication behavior waveform diagram to each first cluster centre distance value the step of.

It also performs the steps of when computer program is executed by processor in one of the embodiments, to Key Performance Null value in achievement data carries out zero filling operation, and deletes the duplicate keys in KPI Key Performance Indicator data.

Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, To any reference of memory, storage, database or other media used in each embodiment provided herein, Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..

Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance Shield all should be considered as described in this specification.

The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.

Claims

1. a kind of construction method of cell scenario category classification model, which comprises the following steps:

Obtain sample cell KPI Key Performance Indicator data, in the KPI Key Performance Indicator data each field data with communicate Behavior field carries out bivariate correlation analysis respectively, removes the field data with communication behavior field correlation lower than threshold value and obtains Obtain communication behavior characteristic；

Communication behavior waveform sample figure is screened from each communication behavior waveform diagram, and obtains each communication behavior waveform sample The corresponding scene type of this figure；

2. the construction method of cell scenario category classification model according to claim 1, which is characterized in that described using each The step of communication behavior characteristic value and binding time granularity acquisition communication behavior waveform diagram, comprising the following steps:

Two-dimensional array is generated using the communication behavior characteristic value and binding time granularity；

It is that ordinate generates communication behavior original image using time granularity as abscissa, communication behavior characteristic value；

The noise in the communication behavior original image is removed, the communication behavior original image after removal noise is determined as communicating Behavior waveform diagram.

3. the construction method of cell scenario category classification model according to claim 1, which is characterized in that described from each institute Screening communication behavior waveform sample figure in communication behavior waveform diagram is stated, and it is corresponding to obtain each communication behavior waveform sample figure The step of scene type, comprising the following steps:

Obtain target scene class number；

According to the target scene class number, quantity is respectively divided in each communication behavior waveform diagram using clustering algorithm To obtain in each scene clustering center and each scene clustering in the different scenes cluster of target scene class number The corresponding scene type of the heart；

The communication behavior waveform diagram of pre-determined distance value will be less than with the distance value at each scene clustering center as current behavior Waveform sample figure；

The corresponding scene type of the current behavior waveform sample figure is determined according to the corresponding scene type in each scene clustering center.

4. the construction method of cell scenario category classification model according to claim 3, which is characterized in that described using poly- Class algorithm each communication behavior waveform diagram is respectively divided in the different scenes cluster that quantity is target scene class number, The step of obtaining each scene clustering center and the corresponding scene type in each scene clustering center, including following step It is rapid:

Randomly selecting quantity is the communication behavior waveform diagram of target scene class number as in the first cluster of each scene clustering The heart；

Calculate remaining communication behavior waveform diagram to each first cluster centre distance value, by each remaining communication behavior wave Shape figure is respectively divided with the first the smallest scene clustering of cluster centre distance value, obtains each communication behavior waveform diagram and is divided into Scene clustering result in different scenes cluster；

The second cluster centre of each scene clustering is calculated according to the scene clustering result, if each second cluster centre and each First cluster centre is equal, then using each second cluster centre as each scene clustering center, and determines that each scene is poly- Class center for scene type.

5. the construction method of cell scenario category classification model according to claim 4, which is characterized in that described according to institute It is further comprising the steps of after stating the step of scene clustering result calculates the second cluster centre of each scene clustering:

If each second cluster centre and each first cluster centre are unequal, using each second cluster centre as first Cluster centre jumps the step of executing the distance value for calculating remaining communication behavior waveform diagram to each first cluster centre.

6. the construction method of cell scenario category classification model according to claim 1, which is characterized in that the acquisition sample It is further comprising the steps of after the step of KPI Key Performance Indicator data of this cell:

Zero filling operation is carried out to the null value in the KPI Key Performance Indicator data, and is deleted in the KPI Key Performance Indicator data Duplicate keys.

7. the construction method of cell scenario category classification model according to claim 1, which is characterized in that the communication row Being characterized data includes network throughput, RRC connection indicated value, up channel Physical Resource Block utilization rate, down channel physics Resource block utilization rate, uplink service information physical resource block occupancy, downlink business information physical resource block occupancy and under At least one of row Successful transmissions initial transmission number of blocks.

8. a kind of construction device of cell scenario category classification model characterized by comprising

Communication behavior characteristic obtains module, for obtaining the KPI Key Performance Indicator data of sample cell, to the crucial achievement Each field data and communication behavior field in effect achievement data carry out bivariate correlation analysis, removal and communication behavior respectively Field correlation obtains communication behavior characteristic lower than the field data of threshold value；

Communication behavior characteristic value acquisition module is led to for carrying out principal component analysis to each communication behavior characteristic Believe behavioural characteristic value；

Communication behavior waveform diagram obtains module, for obtaining communication using each communication behavior characteristic value and binding time granularity Behavior waveform diagram；

Screening sample module for screening communication behavior waveform sample figure from each communication behavior waveform diagram, and obtains each The corresponding scene type of the communication behavior waveform sample figure；

Partitioning model obtains module, for utilizing each communication behavior waveform sample figure and corresponding scene type to depth Learning network model is trained, and the deep learning network model after training is determined as cell scenario category classification model.

9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists In the processor realizes the cell scenario class of any one of claims 1 to 7 the method when executing the computer program The construction step of other partitioning model.

10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program The building of the cell scenario category classification model of method described in any one of claims 1 to 7 is realized when being executed by processor Step.