CN115879029A - Method and device for generating communication network index parameter prediction model - Google Patents

Abstract

The invention discloses a method and a device for generating a communication network index parameter prediction model, which are used for solving the problem that the communication network abnormity is difficult to predict efficiently and accurately. This scheme includes: acquiring historical parameters of a plurality of indexes of a target communication network in a historical period; classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values; determining associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical time period; and taking the classification label corresponding to the parameter value as a training label, taking the correlation characteristic value as a training characteristic value, and training a prediction model based on the historical parameters of the multiple indexes. According to the scheme, effective prediction can be realized before the communication network is abnormal, and the prediction accuracy is improved.

Description

Method and device for generating communication network index parameter prediction model

Technical Field

The invention relates to the field of communication, in particular to a method and a device for generating a communication network index parameter prediction model.

Background

In the field of communication technology, there are many types of causes for causing communication network abnormality, and there are also many types of indexes that reflect communication network states. Technicians often presume the reason of the abnormality according to the connection relation and the index parameters of the communication network after the communication network has failed, and then remove the failure to repair the communication network. The method for repairing the communication network has hysteresis, and the repairing of the communication network usually requires a period of time in which a network user cannot normally use the communication network.

How to predict the communication network abnormality efficiently and accurately is a technical problem to be solved by the application.

Disclosure of Invention

The embodiment of the application aims to provide a method and a device for generating a communication network index parameter prediction model, which are used for solving the problem that the communication network abnormity is difficult to predict efficiently and accurately.

In a first aspect, a method for generating a communication network index parameter prediction model is provided, including:

acquiring historical parameters of a plurality of indexes of a target communication network in a historical period, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values;

classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values;

determining associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical period, wherein the associated characteristic values represent the association relationship among the plurality of indexes, and the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes;

and taking the classification label corresponding to the parameter value as a training label, taking the associated characteristic value as a training characteristic value, training a prediction model based on historical parameters of the multiple indexes, wherein the trained prediction model is used for predicting the parameter of the target index in a second time period according to the input parameter of the target index in the first time period, and the second time period is a time period after the first time period.

In a second aspect, an apparatus for generating a communication network index parameter prediction model is provided, including:

the acquisition module is used for acquiring historical parameters of a plurality of indexes of a target communication network in a historical period, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values;

the classification module is used for classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values;

the determining module is used for determining associated characteristic values of the plurality of indexes according to the variation trends of historical parameters of the plurality of indexes in the historical period, and the associated characteristic values represent the association relation among the plurality of indexes, wherein the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes;

and the training module is used for taking the classification label corresponding to the parameter value as a training label, taking the associated characteristic value as a training characteristic value, training a prediction model based on historical parameters of the plurality of indexes, wherein the trained prediction model is used for predicting the parameter of the target index in a second time period according to the input parameter of the target index in the first time period, and the second time period is a time period after the first time period.

In a third aspect, an electronic device is provided, the electronic device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the method according to the first aspect.

In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, realizes the steps of the method as in the first aspect.

In the embodiment of the application, historical parameters of a plurality of indexes of a target communication network in a historical period are obtained, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values; classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values; determining associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical period, wherein the associated characteristic values represent the association relationship among the plurality of indexes, and the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes; and taking the classification label corresponding to the parameter value as a training label, taking the correlation characteristic value as a training characteristic value, training a prediction model based on historical parameters of the multiple indexes, wherein the trained prediction model is used for predicting parameters of the target index in a second time period according to the input parameters of the target index in a first time period, and the second time period is a time period after the first time period, so that effective prediction can be realized before the occurrence of the abnormity, and the prediction accuracy can be effectively improved by predicting the abnormity of the communication network based on the correlation among the multiple indexes.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:

fig. 1 is a flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 2 is a second flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 3 is a third flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 4 is a flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 5 is a fifth flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 6 is a sixth flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

Fig. 7 is a seventh flowchart illustrating a method for generating a communication network index parameter prediction model according to an embodiment of the present invention.

FIG. 8 is a schematic diagram of an optimization process of an embodiment of the present invention based on a trained model in an application scenario.

Fig. 9 is a schematic structural diagram of a communication network index parameter prediction model generation apparatus according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention. The reference numbers in the present application are only used for distinguishing the steps in the scheme and are not used for limiting the execution sequence of the steps, and the specific execution sequence is described in the specification.

In the technical field of communication, a prediction model can be trained by using historical data of communication indexes, so that the trained prediction model can predict future communication data according to real-time communication data, and further, communication abnormity prediction in a future period is realized. However, various communication indexes are often correlated with each other, and the correlation between different indexes is often different in different states, and if only the various communication indexes are independently predicted, the obtained prediction result is often inaccurate.

In order to solve the problems in the prior art, an embodiment of the present application provides a method for generating a communication network index parameter prediction model, as shown in fig. 1, including:

s11: obtaining historical parameters of a plurality of indexes of a target communication network in a historical period, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values.

In this step, historical parameters of a plurality of indexes of the communication network in a historical period are obtained, the indexes may include, for example, a call ticket establishment success rate, a communication flow rate, and the like, and the indexes may be related to a load of the communication network.

The time length of the historical time period can be two months or longer, and the longer historical time period can reflect the variation trend of the parameters of the multiple indexes and the incidence relation between the multiple indexes more clearly, so that the prediction accuracy of the trained model is improved.

For example, the historical parameter of a certain index may be compared with a preset parameter interval corresponding to the index, and if the historical parameter is within the preset parameter interval range, it is determined that the label corresponding to the historical parameter is a label with normal characteristic data, otherwise, the corresponding label is a label with abnormal characteristic data. It should be understood that preset parameter intervals corresponding to different indexes may be different, and the preset parameter intervals may be manually preset or may be automatically generated according to historical parameters.

S12: classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values;

in this step, the sorting may be performed based on the magnitude of each parameter value, and then the timestamps are classified according to the sorting result, so as to obtain various classified time periods. And then, determining classification labels corresponding to the parameter values based on the time periods obtained by classification. Wherein the classification tag corresponds to a data set comprising a parameter value and a timestamp corresponding to the parameter value.

S13: determining associated characteristic values of the plurality of indexes according to the variation trend of historical parameters of the plurality of indexes in the historical period, wherein the associated characteristic values represent the association relation among the plurality of indexes, and the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes.

There is often an association relationship between multiple indexes in the communication network, for example, under the normal operation condition of the communication network, the call establishment success rate of the wireless base station is associated with the service state and the equipment state of the base station itself. Under the condition that the bearer network has a fault (such as optical cable interruption), the call establishment success rate is associated with the index state of the bearer network. In this step, the correlation characteristic value representing the correlation between the indexes is determined according to the variation trend of the historical parameters of the indexes of the communication network in the historical period. The parameter of the indicator of the call establishment success rate of the wireless base station is the superposition result of the parameters of the two indicators of the service state and the equipment state of the base station.

The associated characteristic values of the multiple indexes can be determined through the step, the associated characteristic values can represent the variation relation of parameters of the associated indexes, and the associated characteristic values are used as a characteristic training model in the subsequent step, so that the accuracy of the model obtained through training for predicting the parameters of the indexes in the future time period can be improved.

S14: and taking the classification label corresponding to the parameter value as a training label, taking the association relation as a training characteristic value, training a prediction model based on historical parameters of the multiple indexes, wherein the trained prediction model is used for predicting parameters of the target index in a second time period according to the input parameters of the target index in the first time period, and the second time period is a time period after the first time period.

The prediction model applied in this step may be determined according to actual requirements, and may be, for example, xgboost (eXtreme Gradient Boosting), gdbt (Gradient Boosting Decision Tree), a neural network, deep learning, or the like. Optionally, the model parameters may also be optimized by using a genetic algorithm to obtain a model with a minimum Mean Absolute Error (MAE). After the training of the prediction model is completed, the parameters of the indexes of the communication network monitored in real time can be input into the trained prediction model, and the output result can represent the prediction parameters of the indexes of the communication network in the future time period.

According to the scheme provided by the embodiment of the application, the incidence relation among the indexes is determined before the prediction model is trained, and then the prediction model is trained based on the incidence relation, so that the trained model can predict the parameters of the indexes in the future period according to the incidence relation among the indexes, and the prediction accuracy is effectively improved.

Because the data of each index in the communication network are mutually influenced, the incidence relation between the indexes is often different under the conditions of busy time and idle time of the communication network. In order to further improve the accuracy of the prediction model obtained by training, the scheme provided by the embodiment of the application performs refined partitioning on the historical parameters so as to perform more accurate correlation analysis on the historical parameters in various time periods respectively.

Based on the solution provided in the foregoing embodiment, optionally, the historical parameter includes a plurality of parameter values corresponding to a plurality of time points in the historical period;

as shown in fig. 2, the step S12 includes:

s21: and sequencing the plurality of parameter values in the historical parameters according to the size relationship to obtain a sequencing result.

For example, the radio bill creation success numbers of the single cell may be sorted, where each parameter value in the sorting result corresponds to a time point in the history period.

S22: and classifying a plurality of time points corresponding to the plurality of parameter values according to the sequencing result.

Each parameter value in the sequencing result can represent the communication network state of the corresponding time point, for example, if the successful creation number of the wireless call ticket is greater than the preset successful creation number, it indicates that the communication network state of the time point corresponding to the successful creation number of the wireless call ticket is a busy hour state, and if the successful creation number of the wireless call ticket is less than or equal to the preset successful creation number, it indicates that the communication network state of the time point corresponding to the successful creation number of the wireless call ticket is a free hour state.

S23: and dividing the historical time interval into a plurality of types of historical sub-time intervals according to the classified plurality of time points.

In general, the communication habits of regional users are generally of a certain popularity, so the state of the communication network has a certain continuity. For example, most users execute the creation of the wireless telephone bill after work, and further the success number of the creation of the wireless telephone bill at 7 o 'clock to 9 o' clock in the evening is often more. Most users have a rest before 12 o ' clock, so that the success rate of creating the wireless telephone bill from 12 o ' clock to 6 o ' clock in the morning is often less. Due to the fact that communication habits of users in different regions are different, the scheme provided by the embodiment of the application classifies corresponding time points according to historical parameters, can obtain idle time and busy time of a communication network by dividing, and further performs correlation analysis aiming at the historical parameters of the communication network in different states, so that the prediction accuracy of the trained prediction model is improved.

It should be understood that in the scheme provided in the embodiment of the present application, a plurality of history sub-periods may be divided according to actual requirements, for example, three types of history sub-periods respectively corresponding to busy time, normal time, and idle time are obtained through division, or a greater number of types of history sub-periods are obtained through division, so as to implement fine division.

In addition, the historical sub-periods may include discontinuous multi-period periods, for example, the busy hour historical sub-periods may include 11 pm to 1 pm and 7 pm to 9 pm. Further, the classified multiple time points can be optimized based on continuity, for example, parameter values corresponding to most time points in the 7-8-point time period are all greater than the preset success number, and only a few of parameter values corresponding to the time points are less than or equal to the preset success number, so that the 7-8-point time period can be divided into busy hour history sub-time periods. Whether to perform the optimization may be determined according to a ratio of a parameter value greater than a preset success number to a parameter value less than or equal to the preset success number, for example, if the ratio of the parameter value greater than the preset success number to the parameter value less than or equal to the preset success number is greater than the preset ratio, the optimization is performed. By the scheme provided by the embodiment of the application, the continuity of the divided historical sub-periods can be improved, and the processing load of the historical parameters is reduced.

Wherein, the step S13 includes:

s24: and respectively determining the associated characteristic values of the multiple indexes in the multiple types of historical sub-periods according to the variation trend of the historical parameters of the multiple indexes in the historical period.

Because the communication network has different loads in busy time and idle time, the incidence relation of each index of the communication network is different in busy time and idle time. According to the scheme provided by the embodiment of the application, the corresponding time points are classified according to the parameter values, and then the associated characteristic values of the indexes are determined according to the parameter change trends in different types of historical sub-periods, so that the prediction accuracy of the prediction model obtained through subsequent training can be improved.

It should be understood that the associated characteristic values of an index in different types of history sub-periods may be different, for example, the call ticket establishment success rate is closely related to the traffic in the idle history sub-period, and is closely related to the index state of the bearer network in the busy history sub-period. In this embodiment, the determined one associated feature value corresponds to the index in one type of history sub-period, and the associated feature values corresponding to the same index in different types of history sub-periods may be different.

Based on the solution provided by the foregoing embodiment, optionally, as shown in fig. 3, the foregoing step S22 includes:

s31: determining at least one quantile of the ranking result.

The quantile is also called quantile, and in the embodiment of the present application, means a numerical value obtained by dividing the sequencing result into several equal parts, for example, a numerical value obtained by dividing the sequencing result into 4 parts is a quartile, and a quantile obtained by dividing the sequencing result into 2 parts is a bipartite, and may also be called a median. Taking the median as an example, the median may be one of the exact middle of all parameter values in the sorting result. If the sorting result has an even number, the average of the two parameter values at the middle can be taken as the median.

S32: and classifying a plurality of time points corresponding to the plurality of parameter values according to the size relationship between the plurality of parameter values and the at least one quantile.

In this step, a plurality of time points corresponding to the parameter values are classified according to the at least one sub-site determined in the above step. For example, based on the size relationship between each parameter value and the quartile point in the sorting result, the parameter values in the sorting result are divided into four equal parts, wherein each part comprises 25% of the parameter values, and then the time points corresponding to the parameter values are also divided into four equal parts. And for the classified time point sets, sorting the time point sets according to the sizes of the parameter values corresponding to the time points in the time point sets, wherein the four time point sets represent the idle time period, the busy time period and the busy time period of the communication network from small to large in sequence.

In practical application, the classified time point sets can be further clustered. For example, for four time point sets sorted from small to large according to the above rule, a time period corresponding to a first time point set is determined as an idle time period, a second time point set is determined as a normal time period, and a third and fourth time point sets are combined to be determined as a busy time period.

It should be noted that, in practical applications, the condition that the parameter value is equal to the quantile point value may be included, and at this time, the parameter value whose value is equal to the quantile point may be classified according to the preset rule. For example, a parameter value having a value equal to a quantile is classified into a category smaller than the quantile. Or dividing the parameter values with the numerical values equal to the quantiles into two parts, wherein the time point corresponding to one part of the parameter values is divided into the categories smaller than the quantile, and the time point corresponding to the other part of the parameter values is divided into the categories larger than the quantile. The division rule of the time point corresponding to the parameter value equal to the quantile point can be preset based on the actual situation.

In addition, in addition to the scheme of performing the division based on the quantile applied in the present embodiment, the time points may be classified based on a normal distribution or other statistical algorithms.

According to the scheme provided by the application, the corresponding time points can be divided into a plurality of classes based on the size relation of the parameter values, so that the division of different state time periods of the communication network is realized, the incidence relation of each index can be respectively determined according to different states of the communication network, and the prediction accuracy of the subsequently generated prediction model is improved.

Based on the solution provided by the foregoing embodiment, optionally, as shown in fig. 4, the foregoing step S24 includes:

s41: and determining the associated characteristics of the multiple indexes in the first type of historical sub-period according to the variation trend of the historical parameters of the multiple indexes in the first type of historical sub-period, wherein the associated characteristics of the indexes represent the characteristic that the parameter of the first index changes along with the parameter of the second index associated with the first index.

In the scheme provided by the embodiment of the application, the associated characteristic values of a plurality of indexes can be determined for various historical sub-periods respectively. Taking the first-type history sub-period as an example, the multiple indexes often show a certain variation trend in the first-type history sub-period, such as gradually rising, gradually falling, increasing in a short time, falling in a short time, and the like. When a parameter of an index changes, one or more indexes associated with the index often change under the influence of the parameter change of the index. In this step, when the parameter of the second index changes, the parameter of the associated first index also changes correspondingly along with the parameter of the second index.

It should be understood that, the parameter of the first index changes along with the parameter of the second index in this step, which means that the parameter of the first index changes due to the parameter of the second index, and the trend and magnitude of the parameter change of the first index may be different from the trend and magnitude of the parameter change of the second index. For example, the parameter of the second index abruptly increases for a short time, and the parameter of the first index slowly decreases after the parameter of the second index increases.

In this step, the correlation characteristic of the index may be specifically determined according to the variation amplitude and variation time period of the parameters of the plurality of indexes in the first-type history sub-period. The parameters of different indexes have different properties, and a part of the indexes have mutation, such as indexes for representing the connection and disconnection of a communication network. There are also some indicators that tend to change slowly, such as indicators of network traffic, traffic volume, etc. The correlation characteristics determined in this step can represent the characteristics of the corresponding index such as the change amplitude, the change time and the like, and the index with parameter mutation and the index with parameter slow change can be distinguished through the correlation characteristics.

S42: and determining the associated characteristic values of the plurality of indexes in the first type of historical sub-period according to the associated characteristics of the plurality of indexes in the first type of historical sub-period.

And further determining the associated characteristic values of the plurality of indexes in the first-class history sub-period based on the associated characteristics of the indexes determined in the steps. For example, the associated characteristic value may indicate which one or more parameters of the second index the parameter of the first index changes, or may indicate the degree of influence of each parameter of the second index on the parameter of the first index. The associated feature values are used for training as a feature input model in a subsequent step.

Through the scheme provided by the embodiment of the application, the correlation characteristics of the indexes can be determined according to the parameter change trend of the indexes. And further determining the associated characteristic value of each index based on the associated characteristic of the index, and improving the accuracy of the associated characteristic value to characterize the index characteristic. And after the correlation characteristic value is used as a training characteristic and input into the model for training, the prediction accuracy of the trained prediction model can be improved.

Based on the scheme provided by the above embodiment, optionally, as shown in fig. 5, the correlation characteristics include a linear correlation characteristic and an abrupt correlation characteristic.

Wherein, the step S41 includes:

s51: and if the change trend of the historical parameter of the first index in the first-class historical sub-period is positively or negatively correlated with the change trend of the historical parameter of the associated second index in the first-class historical sub-period, determining that the correlation characteristic of the first index is a linear correlation characteristic.

In this embodiment, the parameter of the index having the linear correlation characteristic tends to change slowly, and the change of the parameter value has a certain continuity. In this step, the variation trend of the historical parameter of the first indicator in the first type of historical sub-period refers to the overall variation trend of the historical parameter of the first indicator over a period of time, and may include a few fluctuation points that do not meet the overall variation trend. For example, the historical parameter of the first index increases as a whole in a certain period of time, but there are few parameter points in the historical parameter which cause the historical parameter to decrease in a short time, and the parameter value fluctuates. In this case, it is still possible to determine that the change trend of the historical parameter of the first index is a growing trend.

In this step, whether the historical parameter of the first index is positively or negatively correlated with the historical parameter of the second index in the first-type historical sub-period may be determined according to the slope of the change of the historical parameter of the first index and the slope of the change of the historical parameter of the second index in the first-type historical sub-period. For example, the historical parameter of the second index gradually increases, and the increase is larger and larger. The first index changes along with the second index, the historical parameter of the first index is gradually reduced in the corresponding time period, and the reduction amplitude is increased, so that the correlation characteristic of the first index can be determined to be a linear correlation characteristic.

S52: and if the first index does not have the linear correlation characteristic, and the sudden change time period of the historical parameter of the first index in the first type of historical sub-time period is the same as the sudden change time period of the historical parameter of the associated second index in the first type of historical sub-time period, determining that the correlation characteristic of the first index is a sudden change correlation characteristic, wherein the difference between the maximum value and the minimum value of the historical parameter in the sudden change time period is greater than a preset difference.

The time length of the mutation period in this step is not greater than a preset time length, and the preset time length can be manually set in advance. The index with linear correlation characteristic determined in the above step is usually related to the service level of the communication network, while the parameter with abrupt correlation characteristic determined in the step does not change obviously when the parameters of other indexes are increased or decreased slightly, and the other indexes change abruptly correspondingly when they change abruptly. For example, assuming that the first indicator is the call establishment success rate, in a case where the communication network is normal, the traffic of the communication network has no obvious positive correlation or negative correlation with the call establishment success rate. However, when the communication network fails, the traffic of the communication network changes suddenly, and the call establishment success rate also changes suddenly. The step can determine that the correlation characteristic of the index of the call establishment success rate is a sudden change correlation characteristic.

By the scheme provided by the embodiment of the application, the association characteristics of each index in the communication network can be further distinguished, and then association characteristic values are determined according to indexes with different association characteristics. The associated characteristic values can more accurately represent the corresponding indexes, and the prediction model trained by taking the associated characteristic values as training characteristics can more accurately predict the parameters of the indexes.

Based on the solution provided by the foregoing embodiment, optionally, as shown in fig. 6, the foregoing step S42 includes:

s61: if the first index in the first type of historical subinterval has a linear correlation characteristic or a sudden correlation characteristic, determining a correlation characteristic value of the first index according to historical parameters of the first index and the correlated second index in the first type of historical subinterval.

In this step, the respective correlation characteristic values are further determined for the indices having a linear correlation characteristic or a sudden change correlation characteristic. Specifically, it may be determined whether a trend of the parameter of the first index changing along with the parameter of the second index within the first-type history sub-period matches with the associated feature value of the first index. In other words, it is determined that the associated feature values of the first index within the first type of history sub-period have consistency, and then the associated feature value of the first index is further determined.

If the parameter of the first index shows linear correlation characteristics in a part of time interval in the first type of historical sub-time interval and shows abrupt change correlation characteristics in another part of time interval, the correlation characteristics of the index do not have obvious characteristics in the first type of historical sub-time interval, a correlation characteristic value is difficult to determine to represent the change characteristics of the first index in the sub-time interval, and accurate prediction is difficult to realize by a trained prediction model.

Therefore, in the scheme provided by the embodiment of the application, the correlation characteristic value is further determined for the first index with uniform correlation characteristics in the first-class history sub-period. The determined associated characteristic value can accurately represent the change characteristic of the first index in the historical sub-period. And then the associated characteristic value is used as a prediction model obtained by training the training characteristic, so that the parameter of the first index can be predicted more accurately.

Based on the solution provided by the above embodiment, optionally, as shown in fig. 7, before S61, the method further includes:

s71: determining a data characteristic of the first indicator, the data characteristic characterizing a numerical distribution characteristic of data of the indicator.

The data feature described in this step is used to characterize the numerical distribution of the parameter of the first indicator, and may include a continuous feature and a classification feature. For example, the parameters representing the indicators of the user amount and the traffic amount have a certain continuity, i.e. have a continuous nature. While the parameters characterizing the indicators of the on-off status of the communication network may be categorised, e.g. characterizing the network as unavailable with 0 and characterizing the network as available with 1. Here, the parameter of the index having the classification property may be understood as an index having a discontinuous parameter value. For example, if the value range of the parameter of the index is less than or equal to 30 and greater than or equal to 50, the index has classification property.

Wherein, the step S61 includes:

s72: and analyzing the historical parameters of the first index in the first type of historical sub-period by using a preset correlation analysis method corresponding to the data characteristic of the first index.

Indexes with different data characteristics can be analyzed by adopting a proper analysis method so as to improve the accuracy of an analysis result. The parameters of the first index may be formed by stacking parameters of a plurality of second indexes, pairwise correlation analysis may be performed on the plurality of correlated indexes in this step, and the analysis result may indicate the strength of the correlation between the indexes.

Optionally, the relationship between the mutual influences among the devices in the communication network is complex, and the analysis in this step may be performed based on the topological relationship of the communication network. For example, the analysis of this step can be performed according to the data characteristics of the second index Y and the first index X by using the analysis methods shown in the following table:

by the method, the analysis result of the correlation strength between the characterization indexes can be obtained. Optionally, the indexes with correlation strength smaller than the preset strength are removed, and only the indexes with correlation strength are reserved, so that the prediction accuracy of the training model is improved.

S73: and determining the correlation weight of the first index according to the analysis result, wherein the correlation weight represents the correlation strength between the indexes.

Based on the analysis result obtained in the above step, the step further determines the correlation weight of the first index, which is used as a training feature value to represent the correlation strength between the first index and the correlated second index. In general, the value of the association weight may be greater than 0 and less than or equal to 1, and the association weight of the index having the abrupt association characteristic is generally larger. In the case where an index having a mutation-associated property is mutated, the associated index is also often mutated. For example, when a communication network fails, a large number of indicators related to communication data may suddenly change.

Wherein, the step S13 includes:

s74: and taking the label carried by the historical parameter as a training label, taking the associated weight values of the multiple indexes as training characteristic values, and training a prediction model based on the historical parameters of the multiple indexes.

The correlation weight determined in the above steps can represent the correlation between the correlated indexes, and then the model obtained by training with the correlation weight as the characteristic value can predict the index parameters based on the correlation strengths of the correlated indexes, thereby effectively improving the prediction accuracy.

Based on the scheme provided by the above embodiment, it is assumed that the trained model is xgboost, which performs tuning on the model parameters using a genetic algorithm. In practical application, the parameters of the indexes of the communication network in the first time period monitored in real time can be input into the trained model, and the model predicts the parameters of the indexes in the second time period in the future according to the input parameters. And then, comparing the prediction result output by the model with a preset index interval, if the predicted parameter is within the preset index interval, indicating that the parameter is normal, and if the predicted parameter exceeds the preset index interval, indicating that the index is possibly abnormal in a second time interval, and further sending an alarm corresponding to the index which is possibly abnormal in the second time interval, so that a technician can execute a preventive measure in advance aiming at the possibly abnormal index, and effectively avoiding the abnormal index.

FIG. 8 is a schematic diagram of an optimization process of an embodiment of the present invention based on a trained model in an application scenario. The following describes the solution provided in this embodiment with reference to fig. 8. In practical application, the prediction model obtained based on the scheme can acquire the performance index of the communication network in real time, then carry out data preprocessing and performance index calculation on the acquired real-time performance index, and input the relevant parameters obtained by processing calculation into the trained prediction model. And then carrying out abnormal value detection dynamic threshold self-learning based on the trained prediction model, realizing model optimization by updating the abnormal detection model, and executing performance abnormal detection service according to the output result of the model to determine which indexes are possible to be abnormal in the future time period. If the index is possible to be abnormal, outputting abnormal alarm information to an alarm center correspondingly. The alarm center can send feedback data to the model according to the alarm result so as to trigger model optimization. According to the scheme provided by the embodiment of the application, the prediction of future index data can be realized based on the prediction model, the alarm can be given in time to the index which is possibly abnormal, the prediction model can be optimized based on the feedback result, and the automatic prediction and alarm of the index parameters of the communication network can be effectively realized. Optionally, the anomaly detection method can also be used for integrating multiple models to comprehensively detect anomalies by utilizing AI algorithms such as statistical judgment, unsupervised classification and supervised classification and the like through automatic machine learning modeling based on index historical data.

In order to solve the problems in the prior art, an apparatus 90 for generating a communication network index parameter prediction model according to the embodiment of the present application, as shown in fig. 9, includes:

an obtaining module 91, configured to obtain a history parameter of a plurality of indicators of a target communication network in a history period, where the history parameter includes a plurality of parameter values and timestamps corresponding to the parameter values;

a classification module 92, configured to classify the historical parameters according to the parameter values and the corresponding timestamps, so as to obtain classification tags corresponding to the parameter values;

a determining module 93, configured to determine associated feature values of the multiple indexes according to variation trends of historical parameters of the multiple indexes in the historical period, where the associated feature values represent association relationships among the multiple indexes, and a parameter of a first index in the multiple indexes is a superposition result of parameters of multiple second indexes;

the training module 94 is configured to use the classification label corresponding to the parameter value as a training label, use the associated feature value as a training feature value, train a prediction model based on the historical parameters of the multiple indexes, where the trained prediction model is used to predict a parameter of a target index in a second time period according to an input parameter of the target index in a first time period, where the second time period is a time period after the first time period.

According to the device provided by the embodiment of the application, historical parameters of a plurality of indexes of a target communication network in a historical period are obtained, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values; classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values; determining associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical period, wherein the associated characteristic values represent the association relationship among the plurality of indexes, and the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes; and taking the classification label corresponding to the parameter value as a training label, taking the correlation characteristic value as a training characteristic value, training a prediction model based on historical parameters of the multiple indexes, wherein the trained prediction model is used for predicting parameters of the target index in a second time period according to the input parameters of the target index in a first time period, and the second time period is a time period after the first time period, so that effective prediction can be realized before the occurrence of the abnormity, and the prediction accuracy can be effectively improved by predicting the abnormity of the communication network based on the correlation among the multiple indexes.

Preferably, an embodiment of the present invention further provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, and when executed by the processor, the computer program implements each process of the above-mentioned embodiment of the method for generating a communication network index parameter prediction model, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.

The embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the computer program implements each process of the above-mentioned embodiment of the method for generating a communication network indicator parameter prediction model, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.

While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (10)

1. A method for generating a communication network index parameter prediction model is characterized by comprising the following steps:

acquiring historical parameters of a plurality of indexes of a target communication network in a historical period, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values;

classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the historical parameters;

determining associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical period, wherein the associated characteristic values represent the association relationship among the plurality of indexes, and the parameter of a first index in the plurality of indexes is the superposition result of the parameters of a plurality of second indexes;

and taking the classification label corresponding to the parameter value as a training label, taking the associated characteristic value as a training characteristic value, training a prediction model based on historical parameters of the multiple indexes, wherein the trained prediction model is used for predicting the parameter of the target index in a second time period according to the input parameter of the target index in the first time period, and the second time period is a time period after the first time period.

2. The method of claim 1, wherein the historical parameters comprise a plurality of parameter values corresponding to a plurality of time points within the historical period;

classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values, wherein the classification labels comprise:

sorting a plurality of parameter values in the historical parameters according to the size relationship to obtain a sorting result;

classifying a plurality of time points corresponding to the plurality of parameter values according to the sequencing result;

dividing the historical time interval into a plurality of types of historical sub-time intervals according to the classified plurality of time points;

wherein, determining the associated characteristic values of the plurality of indexes according to the variation trend of the historical parameters of the plurality of indexes in the historical period comprises:

and respectively determining the associated characteristic values of the multiple indexes in the multiple types of historical sub-periods according to the variation trends of the historical parameters of the multiple indexes in the historical time period.

3. The method of claim 2, wherein classifying the plurality of time points corresponding to the plurality of parameter values according to the sorting result comprises:

determining at least one quantile of the ranking result;

and classifying a plurality of time points corresponding to the plurality of parameter values according to the size relationship between the plurality of parameter values and the at least one quantile.

4. The method of claim 3, wherein determining the associated preferential treatment of the plurality of indicators within the plurality of types of historical sub-periods respectively according to the variation trend of the historical parameters of the plurality of indicators within the historical period comprises:

determining the associated characteristics of the plurality of indexes in a first type of historical sub-period according to the variation trend of the historical parameters of the plurality of indexes in the first type of historical sub-period, wherein the associated characteristics of the indexes represent the characteristic that the parameter of a first index changes along with the parameter of a second index associated with the first index;

and determining the associated characteristic values of the plurality of indexes in the first type of historical sub-period according to the associated characteristics of the plurality of indexes in the first type of historical sub-period.

5. The method of claim 4, wherein the correlation characteristics include a linear correlation characteristic and an abrupt correlation characteristic;

determining the correlation characteristics of the multiple indexes in a first type of historical sub-period according to the variation trend of the historical parameters of the multiple indexes in the first type of historical sub-period, wherein the determining comprises the following steps:

if the change trend of the historical parameter of the first index in the first-class historical sub-period is positively or negatively correlated with the change trend of the historical parameter of the associated second index in the first-class historical sub-period, determining that the correlation characteristic of the first index is a linear correlation characteristic;

and if the first index does not have the linear correlation characteristic, and the sudden change time period of the historical parameter of the first index in the first type of historical sub-time period is the same as the sudden change time period of the historical parameter of the associated second index in the first type of historical sub-time period, determining that the correlation characteristic of the first index is a sudden change correlation characteristic, wherein the difference between the maximum value and the minimum value of the historical parameter in the sudden change time period is greater than a preset difference.

6. The method of claim 5, wherein determining associated feature values of a plurality of indicators within the first type of historical sub-period based on associated characteristics of the plurality of indicators within the first type of historical sub-period comprises:

if the first index in the first type of historical subinterval has a linear correlation characteristic or a sudden correlation characteristic, determining a correlation characteristic value of the first index according to historical parameters of the first index and the correlated second index in the first type of historical subinterval.

7. The method of claim 6, further comprising, prior to determining the associated characteristic value of the first indicator from historical parameters of the first indicator and associated second indicator within the first type of historical sub-period:

determining a data characteristic of the first indicator, the data characteristic characterizing a numerical distribution characteristic of data of the indicator;

wherein determining the associated characteristic value of the first index according to the historical parameters of the first index and the associated second index in the first type of historical sub-period comprises:

analyzing the historical parameters of the first index in the first type of historical sub-period by using a preset correlation analysis method corresponding to the data characteristic of the first index;

determining a correlation weight of the first index according to an analysis result, wherein the correlation weight represents the correlation strength between indexes;

the training of the prediction model based on the historical parameters of the multiple indexes by using the classification labels corresponding to the parameter values as training labels and the associated characteristic values as training characteristic values comprises the following steps:

and taking the classification label corresponding to the parameter value as a training label, taking the associated weight values of the multiple indexes as training characteristic values, and training a prediction model based on the historical parameters of the multiple indexes.

8. An apparatus for generation of a communication network indicator parameter prediction model, comprising:

the acquisition module is used for acquiring historical parameters of a plurality of indexes of a target communication network in a historical period, wherein the historical parameters comprise a plurality of parameter values and timestamps corresponding to the parameter values;

the classification module is used for classifying the historical parameters according to the parameter values and the corresponding timestamps to obtain classification labels corresponding to the parameter values;

the determining module is used for determining associated characteristic values of the multiple indexes according to the variation trend of historical parameters of the multiple indexes in the historical time period, wherein the associated characteristic values represent the association relation among the multiple indexes, and the parameter of a first index in the multiple indexes is the superposition result of the parameters of multiple second indexes;

and the training module is used for taking the classification label corresponding to the parameter value as a training label, taking the associated characteristic value as a training characteristic value, training a prediction model based on historical parameters of the plurality of indexes, wherein the trained prediction model is used for predicting the parameter of the target index in a second time period according to the input parameter of the target index in a first time period, and the second time period is a time period after the first time period.

9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, which computer program, when executed by the processor, carries out the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, which computer program, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Images