CN109104731B

CN109104731B - Method and device for building cell scene category division model and computer equipment

Info

Publication number: CN109104731B
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: 2021-10-29
Anticipated expiration: 2038-07-04
Also published as: CN109104731A

Abstract

The application relates to a method and a device for building a cell scene classification model, computer equipment and a storage medium. The method comprises the following steps: carrying out bivariate correlation analysis on each field data and communication behavior field in the key performance indicator data of the sample cell, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data; performing principal component analysis on each communication behavior characteristic data to obtain a communication behavior characteristic value; acquiring a communication behavior waveform diagram by utilizing the characteristic values of all communication behaviors and combining time granularity, and screening a communication behavior waveform sample diagram from the communication behavior waveform diagram; and training the deep learning network model by utilizing each communication behavior waveform sample graph and the corresponding scene category thereof, and determining the trained deep learning network model as a cell scene category division model. By adopting the method, the efficiency of synchronous updating of the cell scene categories can be improved, and an important basis is provided for customizing a planning construction scheme and determining a network optimization strategy.

Description

Method and device for building cell scene category division model and computer equipment

Technical Field

The present application relates to the field of communications technologies, and in particular, to a method and an apparatus for constructing a cell scene classification model, a computer device, and a storage medium.

Background

With the continuous development of mobile communication networks, scene division of different cells is an important basis for customizing a planning construction scheme and determining a network optimization strategy.

The division of cell scene categories is determined at the early stage of cell planning construction, the division of the cell scene categories of the traditional mobile communication network is mainly divided artificially by a network planning engineer according to the geographic environment, coverage factors, service characteristics and the like of a cell, but the geographic environment, network conditions and service characteristics of one cell are constantly changed in subsequent use, the cell scene categories also need to be updated synchronously, and a basis is provided for adjusting and optimizing the network.

Disclosure of Invention

Based on this, it is necessary to provide a method and an apparatus for constructing a cell scene classification model, a computer device, and a storage medium for solving the technical problem that it is difficult to quickly re-classify the cell scene classification to achieve update synchronization of the cell scene classification in the conventional cell scene classification technology.

A building method of a cell scene classification model comprises the following steps:

acquiring key performance index data of a sample cell, respectively carrying out bivariate correlation analysis on each field data and a communication behavior field in the key performance index data, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data;

performing principal component analysis on each communication behavior characteristic data to obtain a communication behavior characteristic value;

acquiring a communication behavior oscillogram by utilizing each communication behavior characteristic value and combining time granularity;

screening a communication behavior waveform sample graph from each communication behavior waveform graph, and acquiring a scene type corresponding to each communication behavior waveform sample graph;

and training a deep learning network model by using each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

A building device of a cell scene classification model comprises the following components:

the communication behavior characteristic data acquisition module is used for acquiring key performance index data of a sample cell, carrying out bivariate correlation analysis on each field data and a communication behavior field in the key performance index data, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data;

the communication behavior characteristic value acquisition module is used for performing principal component analysis on each communication behavior characteristic data to acquire a communication behavior characteristic value;

the communication behavior oscillogram acquisition module is used for acquiring a communication behavior oscillogram by utilizing each communication behavior characteristic value and combining time granularity;

the sample screening module is used for screening a communication behavior waveform sample map from each communication behavior waveform map and acquiring a scene type corresponding to each communication behavior waveform sample map;

and the division model acquisition module is used for training the deep learning network model by utilizing each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

A computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:

A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:

According to the building method, device, computer equipment and storage medium of the cell scene category division model, the key performance index data of the cell are obtained, the communication behavior characteristic value is obtained through the data dimensionality reduction of the field relevant to the communication behavior characteristic in the key performance index data, the communication behavior oscillogram of the cell is further obtained, the training sample is screened out from the oscillogram to train the deep learning network model, the key performance index data of the cell is obtained subsequently, the cell can be quickly divided into different scene categories according to the key performance index data, the synchronous updating efficiency of the cell scene categories is improved, and an important basis is provided for customizing a planning construction scheme and determining a network optimization strategy.

Drawings

Fig. 1 is an application environment diagram of a method for constructing a cell scene classification model according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a method for constructing a cell scene classification model according to an embodiment of the present invention;

FIG. 3 is a flow chart of a waveform diagram of communication behavior obtained by utilizing characteristic values of communication behavior in combination with time granularity, in accordance with an embodiment of the present invention;

FIG. 4 is a flowchart illustrating obtaining a waveform sample graph of communication behaviors and corresponding scene types according to an embodiment of the present invention;

FIG. 5 is a flow chart of the present invention for dividing each communication behavior oscillogram into different scene clusters with the number of target scene categories by using a clustering algorithm;

FIG. 6 is a flowchart illustrating a method for constructing a cell scene classification model according to another embodiment of the present invention;

fig. 7 is a diagram of communication waveforms for a cell of the type of an office building in accordance with an embodiment of the present invention;

FIG. 8 is a table of the structure of a deep learning network model in an embodiment of the present invention;

fig. 9 is a schematic structural diagram of a device for constructing a cell scene classification model according to an embodiment of the present invention;

fig. 10 is an internal structural view of a computer device in one embodiment of the present invention.

Detailed Description

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

Referring to fig. 1, fig. 1 is an application environment diagram of a method for constructing a cell scene classification model according to an embodiment of the present invention. In fig. 1, the method for constructing the cell scene classification model may be applied to a system for constructing the cell scene classification model, where the system includes a terminal 110 and a server 120, and the terminal 110 is connected to the server 120 through a network. The terminal 110 may be, but is not limited to, a personal computer, notebook power, tablet computer, etc. The server 120 may obtain the key performance indicator data of the sample cell from the terminal 110, and obtain the communication behavior oscillogram after processing, so as to train the deep learning network model.

Referring to fig. 2, fig. 2 is a flowchart of a method for constructing a cell scene classification model in an embodiment of the present invention, where the method for constructing a cell scene classification model in this embodiment includes the following steps:

step S210: obtaining key performance index data of a sample cell, respectively carrying out bivariate correlation analysis on each field data and a communication behavior field in the key performance index data, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data.

In this step, Key Performance Indicator (KPI) data refers to network management assessment Indicator data of a cell base station, which is used for describing user communication quality parameters; the communication behavior field is field data related to communication behavior in the key performance indicator data, and may be field data such as throughput of a cell base station or communication connection rate.

Specifically, in the network information system, key performance indicator data of a plurality of cell base stations in the whole network can be exported to a server by taking hours as dimensions, after the server obtains the key performance indicator data, the server analyzes the relevance between each field data in the key performance indicator data and a communication behavior field, if the relevance between a certain field data in the key performance indicator data and the communication behavior field is lower than a preset threshold value, the field data is deleted from the key performance indicator data, after the field data in the key performance indicator data, the relevance between the field data and the communication behavior field is lower than the preset threshold value, the remaining field data in the key performance indicator data is used as communication behavior characteristic data. Through bivariate correlation analysis, data clustering is carried out by taking a communication behavior field as a core, and a noise field irrelevant to the communication behavior field is eliminated, so that the interference of the noise field is effectively reduced, the data volume is effectively reduced, and the calculation rate is improved.

Step S220: and performing principal component analysis on each communication behavior characteristic data to obtain a communication behavior characteristic value.

In this step, the communication behavior feature data includes a plurality of field data related to the communication behavior; and the server performs principal component analysis to reduce the dimension by taking the acquired communication behavior characteristic data of a plurality of cell base stations in the whole network as reference data, and obtains a one-dimensional communication behavior characteristic value after reducing the dimension of each multi-dimensional communication behavior characteristic data.

Step S230: and acquiring a communication behavior oscillogram by utilizing the characteristic values of the communication behaviors and combining the time granularity.

In this step, the server combines the obtained communication behavior characteristic value with the time dimension to generate a communication behavior oscillogram related to time. Specifically, the key performance indicator data may be derived by using time as a dimension, after bivariate correlation analysis and principal component analysis operations, a one-dimensional communication behavior feature value may be used to characterize communication characteristics of a cell at a certain time according to the key performance indicator data, and the server may generate a two-dimensional coordinate waveform diagram arranged according to time by using the communication behavior feature value in combination with the time dimension, where the two-dimensional coordinate waveform diagram represents a symbolic communication behavior distribution of the cell.

Step S240: and screening a communication behavior waveform sample diagram from each communication behavior waveform diagram, and acquiring a scene type corresponding to each communication behavior waveform sample diagram.

In this step, a plurality of communication behavior oscillograms with obvious communication behavior distribution characteristics are screened from each communication behavior oscillogram as training sample graphs, and corresponding scene category labels are marked on each communication behavior oscillogram.

Step S250: and training the deep learning network model by utilizing each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

In this step, the server uses each communication behavior waveform sample diagram and the corresponding scene type thereof as training samples to be input into the deep learning network model for supervised training. Specifically, the server may use the communication behavior waveform sample map as an input item of the deep learning network model, and the deep learning network model performs analysis training on the input communication behavior waveform sample map, compares an output recognition result with a scene type corresponding to the communication behavior waveform sample map, and reversely corrects parameters of each layer of network in the deep learning network model according to the comparison result.

Optionally, the communication behavior feature data includes at least one of a network throughput, an RRC connection indication value, an uplink channel physical resource block utilization rate, a downlink channel physical resource block utilization rate, an uplink service information physical resource block occupancy rate, a downlink service information physical resource block occupancy rate, and a number of downlink successful transmission initial transport blocks.

In the method for constructing the cell scene category division model, the key performance indicator data of the cell are obtained, the communication behavior characteristic value is obtained through the dimension reduction of the field data related to the communication behavior characteristic in the key performance indicator data, then the communication behavior oscillogram of the cell is obtained, the training sample is screened out from the oscillogram to train the deep learning network model, the key performance indicator data of the cell is obtained subsequently, the corresponding traffic behavior oscillogram can be obtained from the key performance indicator data, the cell is quickly divided into different scene categories according to the traffic behavior oscillogram, the synchronous updating efficiency of the cell scene categories is improved, and an important basis is provided for customizing a planning construction scheme and determining a network optimization strategy.

Referring to fig. 3, fig. 3 is a flowchart of obtaining a waveform diagram of communication behavior by using characteristic values of communication behavior and combining time granularity according to an embodiment of the present invention; in this embodiment, the step of obtaining the communication behavior oscillogram by using the characteristic values of each communication behavior and combining the time granularity includes the following steps:

step S231: and generating a two-dimensional array by utilizing the communication behavior characteristic value and combining the time granularity.

In this step, the key performance indicator data may be derived by using time as a dimension, after bivariate correlation analysis and principal component analysis operations, a one-dimensional communication behavior feature value may be used to represent communication behavior features of a cell at a certain time according to the key performance indicator data, and the server combines the communication behavior feature value with the corresponding time to generate a two-dimensional array.

Step S232: and generating the original image of the communication behavior by taking the time granularity as an abscissa and the characteristic value of the communication behavior as an ordinate.

In this step, the server combines the communication behavior characteristic values belonging to the same cell and at different times with the time dimension to generate a two-dimensional coordinate oscillogram which takes the time as an abscissa and the communication behavior characteristic values as an ordinate, and the two-dimensional coordinate oscillogram is an original image of the communication behavior.

Step S233: and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

In the step, the noise in the original image of the communication behavior comprises interference factors such as a two-dimensional coordinate axis, an image frame, background color and the like; because the existing picture generation tool generally has interference factors such as coordinate axes and picture frames for the generated picture and can cause noise influence on picture identification, the waveform picture of a pure image is obtained by removing the noise in the original image of the communication behavior. Specifically, interference noise such as coordinate axes and frames in the original image of the communication behavior can be eliminated by utilizing Python programming, a waveform picture of a pure image is obtained, accuracy of the waveform picture of the communication behavior for characterizing the communication behavior is improved, and accuracy of the classification of the cell scene classes is improved when a cell scene class classification model is trained by adopting the waveform picture of the communication behavior in the follow-up process.

In the embodiment, the communication behavior oscillogram is generated to change the multidimensional data with complex data into the visual picture data, so that the communication behavior characteristics of different cells are visualized and digitalized, the data complexity is greatly reduced, the problem of inconsistent multidimensional data dimensions is solved, and meanwhile, the later-stage error correction and updating of the cell scene classification model are facilitated.

Referring to fig. 4, fig. 4 is a flowchart for acquiring waveform sample diagrams of communication behaviors and scene types corresponding to the waveform sample diagrams in an embodiment of the present invention; in this embodiment, the step of screening the communication behavior waveform sample diagrams from the communication behavior waveform diagrams and obtaining the scene types corresponding to the communication behavior waveform sample diagrams includes the following steps:

step S241: and acquiring the number of target scene categories.

In this step, the number of target scene categories refers to the number of scene categories.

Step S242: according to the number of the target scene categories, the clustering algorithm is utilized to divide the communication behavior oscillograms into different scene clusters with the number of the target scene categories, and the scene clustering centers and the scene categories corresponding to the scene clustering centers are obtained.

In this step, after the number of the target scene categories is determined, the communication behavior oscillograms with similar characteristics can be divided into one cluster by using a clustering algorithm, so that each communication behavior oscillogram is divided into different scene clusters with the number of the target scene categories, so as to obtain corresponding scene cluster centers, and the scene categories of each scene cluster are determined according to the communication behavior characteristics.

Step S243: and taking the communication behavior waveform diagram with the distance value from each scene clustering center smaller than the preset distance value as a traffic behavior waveform sample diagram.

In the step, a circle area can be drawn by taking each scene clustering center as a circle center and taking a preset distance value as a radius, the communication behavior oscillograms in each circle area are taken as the traffic behavior oscillogram sample graph, a plurality of communication behavior oscillograms with obvious communication behavior distribution characteristics can be rapidly screened as training samples, and the cost for manually screening the training samples is saved.

Step S244: and determining the scene type corresponding to the traffic behavior waveform sample graph according to the scene type corresponding to each scene clustering center.

In the embodiment, according to the characteristics of the communication behavior oscillograms, the communication behavior oscillograms are divided into a certain number of scene clusters by using a clustering algorithm, the communication behavior oscillograms with obvious graph characteristics close to the scene cluster center are used as training samples, the communication behavior oscillograms with obvious distribution characteristics of a plurality of communication behaviors are quickly screened as the training samples, the cost for manually screening the training samples is saved, the screening of the training samples is more objective and accurate, and the accuracy of the classification of the cell scene is improved.

In one embodiment, the step of obtaining each scene clustering center and the scene category corresponding to each scene clustering center by dividing each communication behavior oscillogram into different scene clusters with the number of target scene categories by using a clustering algorithm includes the following steps: randomly selecting communication behavior oscillograms with the quantity as the number of the target scene categories as a first clustering center of each scene cluster; calculating the distance value from the remaining communication behavior oscillograms to each first clustering center, dividing each remaining communication behavior oscillogram into a scene clustering with the smallest distance value from the first clustering center, and obtaining the scene clustering results of the division of each communication behavior oscillogram into different scene clustering; and calculating a second clustering center of each scene cluster according to the scene clustering result, if each second clustering center is equal to each first clustering center, taking each second clustering center as each scene clustering center, and determining the scene type corresponding to each burn scene clustering center.

In this embodiment, K communication behavior oscillograms are randomly selected from the plurality of communication behavior oscillograms as a first clustering center, where K is the number of target scene categories, and then the distance between each communication behavior oscillogram and the first clustering center is calculated, so that the communication behavior oscillograms are classified into the scene cluster where the first clustering center closest to the communication behavior oscillogram is located. And calculating the average value of the communication behavior oscillogram of each newly formed scene cluster to obtain a second cluster center, and finishing the clustering if the cluster centers of two adjacent scenes do not have any change.

Further, in one embodiment, after the step of calculating the second clustering center of each scene cluster according to the scene clustering result, the method further includes the following steps: and if the second clustering centers are not equal to the first clustering centers, taking the second clustering centers as the first clustering centers, and skipping to execute the step of calculating the distance values from the residual communication behavior oscillograms to the first clustering centers.

In this embodiment, if the cluster centers of two adjacent times change, the second cluster center is regarded as the first cluster center, each communication behavior oscillogram is divided into categories of different cluster centers again, and the cluster centers are calculated iteratively until there is no change in the cluster centers of two adjacent times.

Referring to fig. 5, fig. 5 is a flowchart illustrating that, in an embodiment of the present invention, each communication behavior oscillogram is divided into different scene clusters with the number of target scene categories by using a clustering algorithm, in this embodiment, each communication behavior oscillogram is divided into different scene clusters with the number of target scene categories by using a clustering algorithm, and each scene cluster center and a scene category corresponding to each scene cluster center are obtained, where the method includes the following steps:

step S310: and randomly selecting the communication behavior oscillograms with the quantity as the number of the target scene categories as a first cluster center of each scene cluster.

Step S320: and calculating the distance value from the residual communication behavior oscillograms to each first clustering center, dividing each residual communication behavior oscillogram into the scene clustering with the minimum distance value from the first clustering center, and obtaining the scene clustering result of dividing each communication behavior oscillogram into different scene clustering.

Step S330: calculating a second clustering center of each scene cluster according to the scene clustering result, judging whether the first clustering center is equal to the second clustering center, and if the first clustering center is equal to the second clustering center, jumping to the step S340; if the first cluster center is not equal to the second cluster center, go to step S350.

Step S340: and taking each second clustering center as each scene clustering center, and determining the scene category corresponding to each scene clustering center.

Step S350: the second cluster center is used as the first cluster center, and the process jumps to step S320.

In this embodiment, K communication behavior oscillograms are randomly selected from the plurality of communication behavior oscillograms as a first clustering center, where K is the number of categories of the target scene, and then the distance between each communication behavior oscillogram and the first clustering center is calculated, so as to classify the communication behavior oscillograms into a cluster in which the first clustering center closest to the communication behavior oscillogram is located. And calculating the average value of the newly formed communication behavior oscillogram of each cluster to obtain a second cluster center, finishing the clustering if the cluster centers of two adjacent clusters do not change, and dividing the scene categories of the communication behavior oscillogram by taking the second cluster center as the center again if the cluster centers of two adjacent clusters change so as to ensure the convergence of the clusters and the accuracy of cluster division of the communication behavior oscillogram.

In one embodiment, after the step of obtaining the key performance indicator data of the sample cell, the method further comprises the following steps: and carrying out zero filling operation on null values in the key performance indicator data, and deleting repeated items in the key performance indicator data.

In this embodiment, the key performance indicator data of the sample cell often has the problems that data fields are complete but null values exist, and part of the data has duplicate items identical to the null values, and the like, and the server may perform zero-filling operation on the null values in the obtained key performance indicator data to delete the duplicate items, so as to reduce the noise influence of the data to the greatest extent.

Referring to fig. 6, fig. 6 is a flowchart of a method for constructing a cell scene classification model according to another embodiment of the present invention; in this embodiment, the method for constructing the cell scene classification model includes the following steps:

step S401: and acquiring key performance index data of the sample cell, carrying out zero filling operation on null values in the key performance index data, and deleting repeated items in the key performance index data.

In this step, the server derives annual key performance indicator data of cell base stations of the whole network of a year with hours as dimensions from the system, wherein each key performance indicator data comprises fields: time, Cell name CGI (Cell Global Identifier), antenna angle, Cell bandwidth, operating frequency band, throughput of telecommunication data, average throughput of uplink physical resource blocks, average throughput of downlink physical resource blocks, maximum number of RRC connections, number of RRC connection establishment requests, number of RRC connection establishment successes, average utilization of uplink physical resource blocks, average utilization of downlink physical resource blocks, occupancy of uplink service information physical resource blocks, occupancy of downlink service information physical resource blocks, average uplink interference level of physical resource blocks, number of QPSK mode downlink successful transmission initial transmission blocks, number of 16QAM mode downlink successful transmission initial transmission blocks, number of 64QAM mode downlink successful transmission initial transmission blocks, E-RAB drop rate, wireless drop rate, and the like. After a large amount of sample key performance index data are collected, the server basically cleans the sample key performance index data, zero filling operation is carried out on zero values in the sample key performance index data, repeated items in the sample key performance index data are deleted, and data noise influence is reduced to the maximum extent.

Step S402: and respectively carrying out bivariate correlation analysis on each field data and the communication behavior field in the key performance indicator data, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data.

In this step, the field data in the key performance indicator data is numerous, and the server can perform data clustering on the field data in the key performance indicator data, that is, the noise fields are removed by taking the relevant fields of network throughput, call completing rate and the like and the communication characteristic behaviors of users in a cell as core clusters. Specifically, bivariate correlation analysis is performed on each field data and communication behavior field in the key performance indicator data, field data, such as time, cell name CGI, antenna angle, cell bandwidth, working frequency field and the like in the key performance indicator data, which have correlation with the communication behavior field lower than a threshold value and do not change the cell communication behavior characteristics to a large extent are removed, and the influence of data noise is reduced, wherein the communication behavior field can include field data, such as network throughput, wireless disconnection rate, RRC connection success times and the like.

Step S403: and performing principal component analysis on each communication behavior characteristic data to obtain a communication behavior characteristic value.

In this step, eleven field data with correlation between each two field data being greater than 0.5 threshold are finally screened out as research after removing the field data with low correlation between the key performance indicator data and the communication behavior field, where the eleven field data include network throughput, maximum number of RRC connections, number of RRC connection establishment requests, number of RRC connection establishment successes, utilization rate of uplink channel physical resource blocks, utilization rate of downlink channel physical resource blocks, occupancy rate of uplink service information physical resource blocks, occupancy rate of downlink service information physical resource blocks, number of QPSK mode downlink successful transmission initial transmission blocks, number of 16QAM mode downlink successful transmission initial transmission blocks, and number of 64QAM mode downlink successful transmission initial transmission blocks. And (3) performing principal component analysis by taking communication behavior characteristic data in the key performance index data with hours as dimensionalities of cell base stations of the whole network of a year as reference data to obtain a communication behavior characteristic value.

Step S404: and generating a two-dimensional array by utilizing the communication behavior characteristic value and combining the time granularity.

In this step, the key performance indicator data may be derived by taking hours as dimensions, after bivariate correlation analysis and principal component analysis operations, according to the key performance indicator data, the communication characteristics of a cell at a certain time may be represented by one-dimensional communication behavior characteristic values, and the server combines the communication behavior characteristic values belonging to the same cell at different times on the same day with the corresponding times to generate a two-dimensional array, which is arranged and generated according to a 24-hour sequence.

Step S405: and generating the original image of the communication behavior by taking the time granularity as an abscissa and the characteristic value of the communication behavior as an ordinate.

In this step, the server generates a 24 × 24 two-dimensional coordinate waveform diagram with time as a horizontal coordinate and a communication behavior characteristic value as a vertical coordinate according to the two-dimensional array, and the two-dimensional coordinate waveform diagram is a communication behavior original image, so that communication behavior original images representing communication behavior characteristics of all cells of the whole network of a year every day are obtained.

Step S406: and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

Step S407: and acquiring the number of target scene categories.

In this step, the number of the traditional scene categories is usually from 6 to 13, and on this basis, the clustering algorithm can be used to divide each communication behavior oscillogram into multiple categories of scene categories according to the number of the scene categories from 6 to 13, to obtain clustering results from 6 to 13, and compare indexes such as the intensity of the communication behavior oscillograms in each scene category and the difference between various scene categories in each clustering result, so as to take the scene category with the best overall quality of the clustering results as the target scene category number. Finally, the scene category number category with the best overall quality of the clustering result is 9 categories, and the target scene category number is determined to be 9 categories.

Step S408: according to the number of the target scene categories, the clustering algorithm is utilized to divide the communication behavior oscillograms into different scene clusters with the number of the target scene categories, and the scene clustering centers and the scene categories corresponding to the scene clustering centers are obtained.

In this step, after the communication behavior oscillogram is clustered to obtain the scene clustering centers, the scene category of each scene clustering center is determined according to the communication behavior characteristics of each scene clustering center, and finally nine types of scene categories are defined as: wave type, fault type, shift area office building type, discrete type, business district type, bar type, accommodation district type, and village type. Referring to fig. 7, fig. 7 is a communication waveform diagram of an office building-type cell according to an embodiment of the present invention, in the diagram, the communication behavior between 0 and 8 is less, the communication behavior between 8 and 19 starts to rise and then continues to the peak of the time period, and the communication behavior between 19 and 0 is in a greatly decreasing trend, so as to satisfy the communication behavior characteristics of the office building cell.

Step S409: and taking the communication behavior waveform diagram with the distance value from each scene clustering center smaller than the preset distance value as a traffic behavior waveform sample diagram.

Step S410: and determining the scene type corresponding to the traffic behavior waveform sample graph according to the scene type corresponding to each scene clustering center.

Step S411: and training the deep learning network model by utilizing each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

In this step, after obtaining each communication behavior waveform sample map and its corresponding scene type, inputting the obtained result as a weak supervised training sample to the deep learning network model for supervised training, specifically, each communication behavior waveform sample map is used as an input item of the deep learning network model, the scene type corresponding to each communication behavior waveform sample map is used as an output item of the deep learning network model, the deep learning network model performs analysis training on the input communication behavior waveform sample map, an output recognition result is compared with the scene type corresponding to the communication behavior waveform sample map, and parameters of each layer of network in the deep learning network model are reversely corrected according to the comparison result. The structure of the deep learning network model is shown in fig. 8, the pooling layer in the deep learning network model uses two kinds, namely a maximum pooling layer and a random pooling layer, and the step length of the random pooling layer is set as

The iteration number of the training of the deep learning network model is set to be 100000 times, an Adadelta optimizer is used by the optimizer, the learning rate is set to be 0.001, and the step length is set to be 0.0005. By using the random pooling layer, the fuzzy recognition effect of the deep learning network model is improved, and the robustness of the deep learning network model is improved.

In the embodiment, by obtaining the key performance index data of the cell, the communication behavior characteristic value is obtained through dimensionality reduction of field data related to the communication behavior characteristic in the key performance index data, then a communication behavior waveform sample graph is screened out to train a deep learning network model, and the key performance index data of the cell is obtained subsequently, so that the cell can be rapidly divided into different scene categories according to the key performance index data, the synchronous updating efficiency of the scene categories of the cell is improved, and an important basis is provided for customizing a planning construction scheme and determining a network optimization strategy.

In one embodiment, after the step of determining the trained deep learning network model as the cell scene classification model, the method further includes the following steps: acquiring first key performance index data of a cell to be divided, and acquiring first communication behavior characteristic data of the cell to be divided from the first key performance index data; performing principal component dimensionality reduction on the relevant data of each first communication behavior by using a preset principal component coefficient matrix to obtain a first communication behavior characteristic value of the cell to be divided; generating a first communication behavior oscillogram by utilizing each first communication behavior characteristic value and combining time granularity; and acquiring the scene category of the cell to be divided according to the first communication behavior oscillogram.

In this embodiment, the preset principal component coefficient matrix may be obtained by screening out a one-dimensional element according to a coefficient of principal component analysis when principal component analysis is performed on communication behavior characteristic data of the sample cell; the method comprises the steps that a server obtains key performance index data of a cell to be divided within a certain time period, field data related to communication behaviors are extracted from the key performance index data to serve as communication behavior characteristic data, first communication behavior characteristic data in first key performance index data of the cell to be divided are multiplied by a preset principal component coefficient matrix, dimension reduction is conducted on the first communication behavior characteristic data to obtain a corresponding one-dimensional first communication behavior characteristic value, a communication behavior oscillogram is generated by combining the communication behavior characteristic values with corresponding time, after the communication behavior oscillogram of the cell to be divided is obtained, the graph characteristics of the communication behavior oscillogram are input to a pre-trained deep learning network model, and the scene category of the cell to be divided is obtained. By applying the deep learning network model, the scene types of the cell can be quickly divided under the condition of acquiring the key performance index data of the cell, the updating synchronization of the scene types of the cell is realized when the geographic environment, the network condition and the service characteristics of the cell change in the subsequent use, and then the decision assistance is provided for the optimization and the construction of the cell network. Meanwhile, by acquiring key performance index data of the cell, the communication behavior characteristic value is acquired by dimensionality reduction of field data related to the communication behavior characteristic in the key performance index data, and then a communication behavior oscillogram of the cell is acquired according to the communication behavior characteristic value, and the communication behavior oscillogram is used as a division basis of the scene category of the cell, so that the accuracy of the division of the scene category of the cell is greatly improved, and an important basis is provided for customizing a planning construction scheme and determining a network optimization strategy.

In one embodiment, after the step of obtaining the scene category of the to-be-divided cell output by the cell scene category division model, the network state parameter of the to-be-divided cell may be further adjusted according to the scene category of the to-be-divided cell to optimize the allocation of the cell base station, and the step of adjusting the network state parameter of the to-be-divided cell according to the scene category of the to-be-divided cell specifically includes the following steps: determining a corresponding characteristic service set according to the scene category of the target cell; acquiring quality parameters and reference values of service perception factors of all feature services in a feature service set; determining a perception quality influence factor from the business perception factors according to the quality parameters of the business perception factors and the reference value, and matching target network parameters influencing the perception quality influence factor; and adjusting the target network parameters according to the preset expected performance index of the target cell.

In the embodiment, the situation of key performance index data of each cell base station every day is monitored in real time, the cells with changed scene categories are divided in real time, meanwhile, the feature service set under the scene categories is obtained according to the division result of the scene categories, the perception quality influence factor influencing the use of the feature service experience of a terminal user is determined, the network parameter influencing the perception quality influence factor is adjusted by combining the expected performance index of the cell to be divided, the perception quality influence factor aiming at the feature service under the scene categories of the cell is realized, the key network parameter is accurately adjusted, the service performance of the cell network is effectively improved, the network optimization effect is improved, and the fine adjustment and optimization of the network are realized.

In one embodiment, feature services in the same scene category are the same, and feature services in different scene categories are different, so that a feature service set corresponding to each scene category can be obtained in advance, and the specific steps include: acquiring the access volume and the flow of various services in the full service under the scene type of a target cell; respectively calculating the weight values of various services in the full-volume service according to the access volume and the flow of various services; and determining the service with the weight value ratio exceeding a preset threshold value as the corresponding characteristic service under the scene category of the target cell, and acquiring a characteristic service set.

Specifically, access volume and flow of each type of service in the full-volume service under each type of scene category are counted, based on the difference between the access volume and the flow of each type, the weight value of each type of service under each type of scene category is obtained, and according to the size of the weight value, the service with the weight value ratio exceeding a certain threshold is selected as the feature service of the scene, so that a feature service set is formed. By determining the feature service set of each scene category, when the scene category classification result of the cell is obtained, the feature service set of the cell can be rapidly determined, so that the perception quality influence factor and the corresponding network parameter under the scene category are obtained, and the optimization efficiency of the system is improved.

It should be understood that, although the steps in the flowcharts of fig. 2 to 6 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-6 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performing the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

According to the method for constructing the cell scene classification model, the invention further provides a device for constructing the cell scene classification model, and an embodiment of the device for constructing the cell scene classification model of the invention is described in detail below.

Referring to fig. 9, fig. 9 is a schematic structural diagram of a device for constructing a cell scene classification model according to an embodiment of the present invention, where the device for constructing a cell scene classification model includes: a communication behavior characteristic data obtaining module 510, a communication behavior characteristic value obtaining module 520, a communication behavior oscillogram obtaining module 530, a sample screening module 540, and a division model obtaining module 550, wherein:

a communication behavior feature data obtaining module 510, configured to obtain key performance indicator data of the sample cell, perform bivariate correlation analysis on each field data and a communication behavior field in the key performance indicator data, and remove field data whose correlation with the communication behavior field is lower than a threshold to obtain communication behavior feature data;

a communication behavior characteristic value obtaining module 520, configured to perform principal component analysis on each communication behavior characteristic data to obtain a communication behavior characteristic value;

a communication behavior oscillogram obtaining module 530, configured to obtain a communication behavior oscillogram by using each communication behavior feature value and combining time granularity;

the sample screening module 540 is configured to screen a communication behavior waveform sample map from each communication behavior waveform map, and obtain a scene type corresponding to each communication behavior waveform sample map;

the partitioning model obtaining module 550 is configured to train the deep learning network model by using each communication behavior waveform sample map and the corresponding scene category, and determine the trained deep learning network model as the cell scene category partitioning model.

In one embodiment, the communication behavior oscillogram obtaining module 530 is configured to generate a two-dimensional array by using the communication behavior feature values and combining the time granularity; generating a communication behavior original image by taking the time granularity as an abscissa and the communication behavior characteristic value as an ordinate; and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

In one embodiment, the sample screening module 540 is configured to obtain the number of target scene categories; according to the number of the target scene categories, dividing each communication behavior oscillogram into different scene clusters with the number of the target scene categories by utilizing a clustering algorithm, and acquiring each scene cluster center and a scene category corresponding to each scene cluster center; taking a communication behavior waveform diagram with a distance value smaller than a preset distance value from each scene clustering center as a traffic behavior waveform sample diagram; and determining the scene type corresponding to the traffic behavior waveform sample graph according to the scene type corresponding to each scene clustering center.

In one embodiment, the sample screening module 540 is configured to randomly select the communication behavior oscillograms with the number of the target scene categories as a first cluster center of each scene cluster; calculating the distance value from the remaining communication behavior oscillograms to each first clustering center, dividing each remaining communication behavior oscillogram into a scene clustering with the smallest distance value from the first clustering center, and obtaining the scene clustering results of the division of each communication behavior oscillogram into different scene clustering; and calculating a second clustering center of each scene cluster according to the scene clustering result, taking each second clustering center as each scene clustering center when each second clustering center is equal to each first clustering center, and determining the scene category corresponding to each scene clustering center.

In one embodiment, the sample screening module 540 is configured to take each second clustering center as a first clustering center when each second clustering center is not equal to each first clustering center, and skip the step of calculating the distance value from the remaining communication behavior oscillogram to each first clustering center.

In one embodiment, the communication behavior feature data acquisition module 510 is configured to perform a zero-fill operation on null values in the key performance indicator data and delete duplicate entries in the key performance indicator data.

In one embodiment, the communication behavior characteristic data includes at least one of network throughput, RRC connection indication value, uplink channel physical resource block utilization rate, downlink channel physical resource block utilization rate, uplink traffic information physical resource block occupancy rate, downlink traffic information physical resource block occupancy rate, and downlink successful transmission initial transport block number.

For specific limitations of the device for constructing the cell scene classification model, reference may be made to the above limitations of the method for constructing the cell scene classification model, and details are not repeated here. All or part of each module in the building device of the cell scene classification model can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 10. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing key performance index data of the whole network cell. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method for constructing a cell scene classification model.

Those skilled in the art will appreciate that the architecture shown in fig. 10 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, the present invention further provides a computer device, including a memory and a processor, where the memory stores a computer program, and the processor implements the following steps when executing the computer program:

acquiring a communication behavior oscillogram by utilizing the characteristic values of all communication behaviors and combining time granularity;

screening a communication behavior waveform sample diagram from each communication behavior waveform diagram, and acquiring a scene type corresponding to each communication behavior waveform sample diagram;

and training the deep learning network model by utilizing each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

In one embodiment, when the processor executes the computer program to implement the step of obtaining the waveform diagram of the communication behavior by using the characteristic values of the communication behavior and combining the time granularity, the following steps are specifically implemented: generating a two-dimensional array by utilizing the communication behavior characteristic value and combining the time granularity; generating a communication behavior original image by taking the time granularity as an abscissa and the communication behavior characteristic value as an ordinate; and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

In one embodiment, when the processor executes the computer program to realize the steps of screening the communication behavior waveform sample diagrams from the communication behavior waveform diagrams and acquiring the scene types corresponding to the communication behavior waveform sample diagrams, the following steps are specifically realized: acquiring the number of target scene categories; according to the number of the target scene categories, dividing each communication behavior oscillogram into different scene clusters with the number of the target scene categories by utilizing a clustering algorithm, and acquiring each scene cluster center and a scene category corresponding to each scene cluster center; taking a communication behavior waveform diagram with a distance value smaller than a preset distance value from each scene clustering center as a traffic behavior waveform sample diagram; and determining the scene type corresponding to the traffic behavior waveform sample graph according to the scene type corresponding to each scene clustering center.

In one embodiment, the processor executes a computer program to implement the steps of using a clustering algorithm to divide each communication behavior oscillogram into different scene clusters with the number of target scene categories, and obtaining each scene cluster center and a scene category corresponding to each scene cluster center, and specifically implementing the following steps: randomly selecting communication behavior oscillograms with the quantity as the number of the target scene categories as a first clustering center of each scene cluster; calculating the distance value from the remaining communication behavior oscillograms to each first clustering center, dividing each remaining communication behavior oscillogram into a scene clustering with the smallest distance value from the first clustering center, and obtaining the scene clustering results of the division of each communication behavior oscillogram into different scene clustering; and calculating a second clustering center of each scene cluster according to the scene clustering result, and if each second clustering center is equal to each first clustering center, taking each second clustering center as each scene clustering center, and determining the scene category corresponding to each scene clustering center.

In one embodiment, the processor, when executing the computer program, further performs the steps of: and if the second clustering centers are not equal to the first clustering centers, taking the second clustering centers as the first clustering centers, and skipping to execute the step of calculating the distance values from the residual communication behavior oscillograms to the first clustering centers.

In one embodiment, the processor, when executing the computer program, further performs the steps of: and carrying out zero filling operation on null values in the key performance indicator data, and deleting repeated items in the key performance indicator data.

In one embodiment, the present invention further provides a computer readable storage medium having a computer program stored thereon, the computer program when executed by a processor implementing the steps of:

In one embodiment, when the computer program is executed by the processor to implement the step of obtaining the waveform diagram of the communication behavior by using the characteristic values of the communication behavior and combining the time granularity, the following steps are specifically implemented: generating a two-dimensional array by utilizing the communication behavior characteristic value and combining the time granularity; generating a communication behavior original image by taking the time granularity as an abscissa and the communication behavior characteristic value as an ordinate; and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

In one embodiment, when the computer program is executed by the processor to implement the steps of screening the communication behavior waveform sample diagrams from the communication behavior waveform diagrams and obtaining the scene types corresponding to the communication behavior waveform sample diagrams, the following steps are specifically implemented: acquiring the number of target scene categories; according to the number of the target scene categories, dividing each communication behavior oscillogram into different scene clusters with the number of the target scene categories by utilizing a clustering algorithm, and acquiring each scene cluster center and a scene category corresponding to each scene cluster center; taking a communication behavior waveform diagram with a distance value smaller than a preset distance value from each scene clustering center as a traffic behavior waveform sample diagram; and determining the scene type corresponding to the traffic behavior waveform sample graph according to the scene type corresponding to each scene clustering center.

In one embodiment, when the computer program is executed by the processor to implement the steps of dividing each communication behavior oscillogram into different scene clusters with the number of target scene classes by using a clustering algorithm, and obtaining each scene cluster center and a scene class corresponding to each scene cluster center, the following steps are specifically implemented: randomly selecting communication behavior oscillograms with the quantity as the number of the target scene categories as a first clustering center of each scene cluster; calculating the distance value from the remaining communication behavior oscillograms to each first clustering center, dividing each remaining communication behavior oscillogram into a scene clustering with the smallest distance value from the first clustering center, and obtaining the scene clustering results of the division of each communication behavior oscillogram into different scene clustering; and calculating a second clustering center of each scene cluster according to the scene clustering result, and if each second clustering center is equal to each first clustering center, taking each second clustering center as each scene clustering center, and determining the scene category corresponding to each scene clustering center.

In one embodiment, the computer program when executed by the processor further performs the steps of: and if the second clustering centers are not equal to the first clustering centers, taking the second clustering centers as the first clustering centers, and skipping to execute the step of calculating the distance values from the residual communication behavior oscillograms to the first clustering centers.

In one embodiment, the computer program when executed by the processor further performs the steps of: and carrying out zero filling operation on null values in the key performance indicator data, and deleting repeated items in the key performance indicator data.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A method for constructing a cell scene classification model is characterized by comprising the following steps:

acquiring key performance index data derived by time dimension of a sample cell, respectively carrying out bivariate correlation analysis on each field data and a communication behavior field in the key performance index data, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data;

acquiring the number of target scene categories; according to the number of the target scene categories, the communication behavior oscillogram is divided into different scene clusters with the number of the target scene categories by utilizing a clustering algorithm, and each scene cluster center and a scene category corresponding to each scene cluster center are obtained;

taking the communication behavior waveform diagram with the distance value between the communication behavior waveform diagram and each scene clustering center smaller than a preset distance value as a communication behavior waveform sample diagram;

determining scene types corresponding to the communication behavior waveform sample graph according to the scene types corresponding to the scene clustering centers; and training a deep learning network model by using each communication behavior waveform sample graph and the corresponding scene category, and determining the trained deep learning network model as a cell scene category division model.

2. The method for constructing the cell scene classification model according to claim 1, wherein the step of obtaining the communication behavior oscillogram by using each of the communication behavior feature values and combining the time granularity includes the steps of:

generating a two-dimensional array by utilizing the communication behavior characteristic value and combining time granularity;

generating a communication behavior original image by taking the time granularity as an abscissa and the communication behavior characteristic value as an ordinate;

and removing noise in the original image of the communication behavior, and determining the original image of the communication behavior after the noise is removed as a waveform diagram of the communication behavior.

3. The method for constructing the cell scene classification model according to claim 1, wherein the step of obtaining the scene clustering centers and the scene classifications corresponding to the scene clustering centers by using a clustering algorithm to classify the communication behavior oscillograms into different scene clusters with the number of target scene classifications comprises the following steps:

randomly selecting communication behavior oscillograms with the quantity as the number of the target scene categories as a first clustering center of each scene cluster;

calculating the distance value from the remaining communication behavior oscillograms to each first clustering center, dividing each remaining communication behavior oscillogram into a scene clustering with the smallest distance value from the first clustering center, and obtaining the scene clustering results of the division of each communication behavior oscillogram into different scene clustering;

and calculating second clustering centers of the scene clusters according to the scene clustering results, if the second clustering centers are equal to the first clustering centers, taking the second clustering centers as the scene clustering centers, and determining the scene category corresponding to the scene clustering centers.

4. The method for constructing a cell scene classification model according to claim 3, wherein after the step of calculating the second clustering center of each scene cluster according to the scene clustering result, the method further comprises the following steps:

and if the second clustering centers are not equal to the first clustering centers, taking the second clustering centers as the first clustering centers, and skipping to execute the step of calculating the distance values from the residual communication behavior oscillograms to the first clustering centers.

5. The method for constructing the cell scene classification model according to claim 1, wherein the step of obtaining the key performance indicator data of the sample cell further comprises the following steps:

and carrying out zero filling operation on null values in the key performance indicator data, and deleting repeated items in the key performance indicator data.

6. The method for constructing a cell scene classification model according to claim 1, wherein the communication behavior feature data includes at least one of network throughput, RRC connection indication value, uplink channel physical resource block utilization rate, downlink channel physical resource block utilization rate, uplink service information physical resource block occupancy rate, downlink service information physical resource block occupancy rate, and number of downlink successful transmission initial transport blocks.

7. A building device of a cell scene classification model is characterized by comprising the following components:

the communication behavior characteristic data acquisition module is used for acquiring key performance index data derived by time dimension of a sample cell, carrying out bivariate correlation analysis on each field data and a communication behavior field in the key performance index data respectively, and removing the field data of which the correlation with the communication behavior field is lower than a threshold value to obtain communication behavior characteristic data;

the sample screening module is used for acquiring the number of target scene categories; according to the number of the target scene categories, the communication behavior oscillogram is divided into different scene clusters with the number of the target scene categories by utilizing a clustering algorithm, and each scene cluster center and a scene category corresponding to each scene cluster center are obtained;

determining scene types corresponding to the communication behavior waveform sample graph according to the scene types corresponding to the scene clustering centers;

8. The building device of the cell scene classification model according to claim 7, wherein the sample screening module is configured to randomly select a communication behavior oscillogram whose number is the number of the target scene classes as a first clustering center of each scene cluster;

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the building steps of a cell scene classification model according to the method of any of claims 1 to 6.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the building steps of a cell scene classification model of the method of any one of claims 1 to 6.