CN114726745B - Network traffic prediction method, device and computer readable storage medium - Google Patents

Abstract

The application discloses a network traffic prediction method, a device and a computer readable storage medium, relating to the technical field of artificial intelligence, wherein the method comprises the following steps: acquiring historical flow data of each region in a target range, wherein the historical flow data comprises first flow data corresponding to a region to be predicted; determining a feature matrix corresponding to the region to be predicted based on the first flow data; inputting the historical flow data and the feature matrix into a preset model, and determining a flow prediction result of the area to be predicted; the preset models comprise a first preset model, a second preset model and a third preset model, wherein the first preset model is respectively cascaded with the second preset model and the third preset model, and the second preset model is cascaded with the third preset model. In this way, the flow of the area to be predicted can be predicted through a plurality of models with cascade relations, so that the influence of the feature matrix of the area to be predicted and the flow of other areas in the target range is considered by the prediction result, and the accuracy of the prediction result is improved.

Description

Network traffic prediction method, device and computer readable storage medium

Technical Field

The present invention relates to the field of artificial intelligence technology, and in particular to a network traffic prediction method, device and computer-readable storage medium.

Background technique

With the development of the fourth generation mobile communication network (ie 4G) and the fifth generation mobile communication network (ie 5G), a large number of network services have been integrated into people's lives through various terminal devices, and people's requirements for network speed and latency have gradually increased. In order to ensure a good user experience, it is necessary to ensure that the network is given enough resources. Operators usually formulate network resource allocation strategies based on provincial companies. Network experts in each provincial company need to determine the network congestion status of each city in the future by predicting network traffic based on the time series data of network performance indicators in various cities in the province, and formulate reasonable resource allocation plans based on the resource load of each subordinate city, so as to meet the network resource needs of various cities while maximizing the avoidance of resource waste.

However, the existing network traffic prediction methods usually split the original time series into multiple sub-time series, predict the multiple sub-time series separately, and finally get the prediction result of the original time series by adding the prediction results of each sub-time series. Since this method only fits the prediction result from a mathematical perspective, it is suitable for the prediction of a single time series that is not affected by other time series. However, when applied to the network traffic prediction scenario, it will ignore the similarity and correlation of traffic changes between cities in the same province, resulting in a low accuracy rate of the prediction results.

Summary of the invention

The embodiments of the present invention provide a network traffic prediction method, device and computer-readable storage medium to solve the problem of low accuracy of prediction results of existing network traffic prediction methods.

To solve the above technical problems, the present invention is achieved as follows:

In a first aspect, an embodiment of the present invention provides a network traffic prediction method, the method comprising:

Acquire historical traffic data of each area within the target range, wherein the historical traffic data includes first traffic data corresponding to the area to be predicted;

Based on the first traffic data, determining a feature matrix corresponding to the area to be predicted;

Inputting the historical traffic data and the characteristic matrix into a preset model to determine the traffic prediction result of the area to be predicted;

The preset model includes a first preset model, a second preset model and a third preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model.

Optionally, inputting the historical traffic data and the feature matrix into a preset model to determine the traffic prediction result of the area to be predicted includes:

Inputting the first flow data and the feature matrix into the first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence;

Inputting the initial flow prediction sequence and the feature matrix into the second preset model to obtain a second prediction result, wherein the second prediction result includes a flow difference feature sequence;

Inputting the first prediction result and the flow difference characteristic sequence into the third preset model to obtain the flow prediction result of the area to be predicted, wherein the flow difference characteristic sequence is used to indicate the influence of the population flow of each area within the target range on the flow of the area to be predicted;

Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model, and the third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence.

Optionally, the first prediction result further includes:

At least one of the holiday sequence, trend sequence and season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction time period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction time period, and the season sequence is used to indicate the season characteristics of the area to be predicted within the preset prediction time period.

Optionally, the second preset model includes N sub-models, the second prediction result includes N second prediction sub-results, and N is a positive integer;

The step of inputting the initial flow prediction sequence and the feature matrix into a second preset model to obtain a second prediction result includes:

Inputting the initial traffic prediction sequence and the feature matrix into the target sub-model to obtain a second prediction sub-result;

The target sub-model is any sub-module among the N sub-models.

Optionally, before inputting the initial traffic prediction sequence and the feature matrix into a second preset model to obtain a second prediction result, the method further includes:

Performing training and learning on the historical traffic data and the feature matrix to obtain the N sub-models;

Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow type, outflow type and stable type.

Optionally, the training and learning of the historical traffic data and the feature matrix to obtain the N sub-models includes:

Classify the historical flow data of each area according to the population flow type of each area within the target range;

Calculate N mean sequences corresponding to the N population flow types respectively, where the mean sequences are used to indicate the flow mean values corresponding to multiple areas with the same population flow type within the target range at various collection time points;

Based on the first flow data and the N mean sequences, determine N difference feature sequences corresponding to the N population flow types, the difference feature sequences comprising differences between flow values corresponding to each acquisition time point in the first flow data and flow means at corresponding time points in the mean sequence;

The first traffic data and the feature matrix are used as inputs of each of the N sub-models, and the N differential feature sequences are used as outputs of each of the N sub-models, and the N sub-models are obtained through training and learning.

Optionally, determining the feature matrix corresponding to the to-be-predicted area based on the first traffic data includes:

Performing feature marking on the first flow data to obtain a plurality of feature sequences;

Based on the multiple feature sequences, determining a feature matrix corresponding to the area to be predicted;

Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values, holiday information and activity information corresponding to the multiple collection time points, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences.

In a second aspect, an embodiment of the present invention provides a network traffic prediction device, the device comprising:

An acquisition module, used to acquire historical traffic data of each area within the target range, wherein the historical traffic data includes first traffic data corresponding to the area to be predicted;

A first determination module, configured to determine a feature matrix corresponding to the area to be predicted based on the first traffic data;

A second determination module is used to input the historical traffic data and the characteristic matrix into a preset model to determine the traffic prediction result of the area to be predicted;

The preset model includes a first preset model, a second preset model and a third preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model.

Optionally, the second determining module includes:

A first input submodule, used for inputting the first flow data and the feature matrix into the first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence;

A second input submodule, used for inputting the initial flow prediction sequence and the feature matrix into the second preset model to obtain a second prediction result, wherein the second prediction result includes a flow difference feature sequence;

A third input submodule is used to input the first prediction result and the flow difference feature sequence into the third preset model to obtain the flow prediction result of the area to be predicted, and the flow difference feature sequence is used to indicate the impact of the population flow of each area within the target range on the flow of the area to be predicted;

Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model, and the third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence.

Optionally, the first prediction result further includes:

At least one of the holiday sequence, trend sequence and season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction time period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction time period, and the season sequence is used to indicate the season characteristics of the area to be predicted within the preset prediction time period.

Optionally, the second preset model includes N sub-models, the second prediction result includes N second prediction sub-results, and N is a positive integer; the second input submodule includes:

An input unit, used for inputting the initial traffic prediction sequence and the feature matrix into the target sub-model to obtain a second prediction sub-result;

Among them, the target sub-model is any sub-module among the N sub-models.

Optionally, the second determining module further includes:

A training and learning submodule, used for training and learning the historical traffic data and the feature matrix to obtain the N submodels;

Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow type, outflow type and stable type.

Optionally, the training and learning submodule is specifically used to:

Classify the historical flow data of each area according to the population flow type of each area within the target range;

Calculate N mean sequences corresponding to the N population flow types respectively, where the mean sequences are used to indicate the flow mean values corresponding to multiple areas with the same population flow type within the target range at various collection time points;

Based on the first flow data and the N mean sequences, determine N difference feature sequences corresponding to the N population flow types, the difference feature sequences comprising differences between flow values corresponding to each acquisition time point in the first flow data and flow means at corresponding time points in the mean sequence;

The first traffic data and the feature matrix are used as inputs of each of the N sub-models, and the N differential feature sequences are used as outputs of each of the N sub-models, and the N sub-models are obtained through training and learning.

Optionally, the first determining module includes:

A marking submodule, used for performing feature marking on the first flow data to obtain a plurality of feature sequences;

A second determination submodule is used to determine a feature matrix corresponding to the area to be predicted based on the multiple feature sequences;

Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values, holiday information and activity information corresponding to the multiple collection time points, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences.

In a third aspect, an embodiment of the present invention further provides a network traffic prediction device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method described in the first aspect.

In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the method described in the first aspect are implemented.

In an embodiment of the present invention, by acquiring historical traffic data of each area within the target range, the historical traffic data includes first traffic data corresponding to the area to be predicted; based on the first traffic data, determining the feature matrix corresponding to the area to be predicted; inputting the historical traffic data and the feature matrix into a preset model to determine the traffic prediction result of the area to be predicted; wherein the preset model includes a first preset model, a second preset model and a third preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model. In this way, the network traffic of the area to be predicted can be predicted by multiple models with a cascade relationship, so that the prediction result takes into account the feature matrix of the area to be predicted and the traffic influence of other areas within the target range, thereby improving the accuracy of the prediction result.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

FIG1 is a flow chart of a network traffic prediction method according to an embodiment of the present invention;

FIG2 is a second flowchart of a network traffic prediction method provided by an embodiment of the present invention;

3 is a flowchart of a third method for predicting network traffic according to an embodiment of the present invention;

FIG4 is a schematic diagram of training data provided by an embodiment of the present invention;

FIG5 is a schematic diagram of the structure of a preset model provided in an embodiment of the present invention;

FIG6 is a schematic diagram of a network traffic prediction device according to an embodiment of the present invention;

FIG. 7 is a second schematic diagram of the structure of the network traffic prediction device provided in an embodiment of the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

In the embodiments of the present invention, a network traffic prediction method, apparatus and related equipment are proposed to solve the problem that the existing network traffic prediction method only fits the prediction result from a mathematical perspective, is suitable for a single time series prediction that is not affected by other time series, and ignores the similarity and correlation of traffic changes between cities in the same province, resulting in a low accuracy rate of the prediction result.

Referring to Figure 1, Figure 1 is a flow chart of a network traffic prediction method provided by an embodiment of the present invention. As shown in Figure 1, the method includes the following steps:

Step 101: Acquire historical traffic data of each area within a target range, where the historical traffic data includes first traffic data corresponding to the area to be predicted.

The target range may include multiple regions, and the target range may be the network coverage range corresponding to a certain country, or the network coverage range of a certain province in a certain country, or the network coverage range of a certain city in a certain province, etc. When the target range is the network coverage range corresponding to a certain country, the region indicates the network coverage range of a certain province in the country; when the target range is the network coverage range of a certain province in a certain country, the region indicates the network coverage range of a certain city in the province; when the target range is the network coverage range of a certain city in a certain province, the region indicates the network coverage range of a certain district or county in the city.

The above historical traffic data is the data collected by each area within the target range within the historical preset time. The historical preset time here can be any time such as half a year, one year, two years, three years, etc., and the collected time granularity can be any granularity such as hours, days, months, etc., which is not specifically limited in this application. The above historical traffic data can include data of multiple dimensions, such as collection time, traffic value corresponding to each collection time, holiday information corresponding to each collection time, activity information corresponding to each collection time, etc. For example, assuming that the target range includes 10 areas, by collecting data of four dimensions of each collection time, traffic value corresponding to each collection time, holiday information corresponding to each collection time, and activity information corresponding to each collection time in these 10 areas within the historical preset time, 10 time series {W} _w×4 can be obtained, where {W} represents the four-dimensional time series corresponding to w time collection points in a certain area, w represents the length of the time series, 4 represents the dimension of the time series, and the number of w is determined by the historical preset time and the collected time granularity.

Step 102: Determine a feature matrix corresponding to the area to be predicted based on the first traffic data.

The first traffic data may include data of multiple dimensions collected in the area to be predicted within a preset historical time period, such as collection time, traffic value corresponding to each collection time, holiday information corresponding to each collection time, activity information corresponding to each collection time, etc. The feature matrix includes multiple feature sequences, each feature sequence is used to indicate a feature of the first traffic data, such as time feature, holiday feature, activity feature, area feature, etc.

In one embodiment, the first traffic data includes the collection time, the traffic value corresponding to each collection time, the holiday information corresponding to each collection time, and the activity information corresponding to each collection time, which is represented by {W1} _w×4 , where W1 represents the four-dimensional data matrix corresponding to the w time collection points in the area to be predicted, and the number of w is determined by the historical preset time length and the time granularity collected. In this way, the first traffic data can be marked with features through big data analysis, and its features include time features, holiday features, activity features and regional features, thereby obtaining a time feature sequence, a holiday feature sequence, an activity feature sequence and a regional feature sequence, and then forming a feature matrix {F} _w×k according to the time feature sequence, the holiday feature sequence, the activity feature sequence and the regional feature sequence, where {F} represents a set of k-dimensional feature sequences corresponding to multiple w time collection points, w represents the length of the feature sequence, k represents the dimension of the feature sequence, the number of w is determined by the historical preset time length and the time granularity collected, and k is an arbitrary positive integer. Specifically, the time feature is used to indicate the time information of each collection time of the first flow data, such as the corresponding season (i.e., spring, summer, autumn or winter), the corresponding time period (i.e., morning, afternoon, noon or evening), whether it is a rush hour, etc.; the holiday feature is used to indicate the holiday information of each collection time of the first flow data, such as whether it is a holiday, the name of the holiday, the duration of the holiday, etc.; the activity feature is used to indicate the activity information of each collection time of the first flow data, such as whether it is an activity day, the name of the activity, the duration of the activity, the scope of the activity area, etc.; the regional feature is used to indicate the information of the area to be predicted, such as the type of population flow (i.e., population inflow type, population outflow type or population stability type), the type of city (i.e., tourist city, agricultural city or industrial city), the gross domestic product (GDP) value of the month, the number of permanent residents in the region, the number of urban population, the number of colleges and universities, the number of primary and secondary schools, etc.

Step 103: Input the historical traffic data and the feature matrix into the preset model to determine the traffic prediction result of the area to be predicted.

The preset model may include multiple trained models, which can predict the network traffic of the predicted area within a preset time period in the future according to the input historical traffic data and feature matrix to obtain the traffic prediction result. Specifically, the preset model may include a first preset model for predicting the initial traffic prediction sequence, holiday sequence, trend sequence and seasonal sequence of the predicted area, a second preset model for predicting the traffic conversion relationship between the predicted area and other areas within the target range, and a third preset model for correcting the initial traffic prediction sequence, wherein the first preset model may be a machine learning model such as a prophet model, which is used to obtain the periodic characteristics of the predicted area, the second preset model may be a machine learning model such as a long short-term memory network (Long Short-Term Memory, referred to as LSTM) model, which is used to obtain the time density characteristics of the predicted area, and the third preset model may be an extreme gradient boosting (eXtreme Gradient Boosting, referred to as XGBoost) model, etc. The preset model includes a first preset model, a second preset model and a third preset model. The first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model. In this way, the prediction result of the first preset model can be used as input data for the second preset model and the third preset model, and the prediction result of the second preset model can be used as input data for the third preset model. Finally, the traffic prediction result of the area to be predicted is determined by the third preset model.

In this embodiment, the network traffic of the area to be predicted can be predicted by multiple models with a cascade relationship, so that the prediction result takes into account the feature matrix of the area to be predicted and the traffic influence of other areas within the target range, thereby improving the accuracy of the prediction result.

Further, referring to FIG. 2 , FIG. 2 is a second flowchart of a network traffic prediction method provided by an embodiment of the present invention. Based on the embodiment shown in FIG. 1 , the above step 103, inputting historical traffic data and a feature matrix into a preset model to determine a traffic prediction result of a to-be-predicted area, includes:

Step 201: input first flow data and a feature matrix into a first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence.

In one embodiment, the preset model includes a first preset model, a second preset model and a third preset model, the first preset model and the second preset model are different machine learning models, the third preset model is a correction model, and the third preset model is used to correct the initial traffic prediction sequence in the first prediction result according to the second prediction result. In the preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model.

Specifically, the above-mentioned first preset model is obtained by pre-training and learning the first historical traffic data of the area to be predicted. The first historical traffic data here can be the same as the above-mentioned first traffic data, or it can be different from the above-mentioned first traffic data. In order to make the prediction of the first preset model obtained by training more accurate, historical traffic data for a longer period of time can be selected as training data, such as historical traffic data for the past 5 years. Through these 5 years of historical traffic data, the initial traffic prediction sequence, holiday sequence, trend sequence and seasonal sequence of the predicted area within the preset prediction period (such as within the next 1 year) are obtained. The above-mentioned first preset model includes but is not limited to the prophet model. The specific training and learning process of the first preset model is a prior art and will not be repeated here.

When making predictions through the first preset model, the first traffic data and the feature matrix can be input into the first preset model, thereby outputting a first prediction result, which includes but is not limited to an initial traffic prediction sequence, a holiday sequence, a trend sequence, and a seasonal sequence.

Step 202: Input the initial traffic prediction sequence and the feature matrix into a second preset model to obtain a second prediction result, wherein the second prediction result includes a traffic difference feature sequence, and the traffic difference feature sequence is used to indicate the impact of population flow in each area within the target range on the traffic in the predicted area.

Specifically, the above-mentioned second preset model includes one or more neural network models corresponding to the population flow type, and the neural network model includes but is not limited to a long short-term memory network (Long Short-Term Memory, referred to as LSTM) model.

When the second preset model is used for prediction, the initial flow prediction sequence and the feature matrix can be input into the corresponding neural network model to obtain the second preset result. The second preset result includes one or more flow difference feature sequences corresponding to the population flow type, where the flow difference feature sequence is used to indicate the impact of the population flow in each area within the target range on the flow of the area to be predicted.

Step 203, input the first prediction result and the traffic difference feature sequence into the third preset model to obtain the traffic prediction result of the area to be predicted. Specifically, the above-mentioned third preset model includes but is not limited to the extreme gradient boosting (eXtremeGradient Boosting, referred to as XGBoost) model, which is not specifically limited in this application. When training the third model, it is necessary to use the output results of the first preset model and the second preset model as input data of the third preset model for training and learning, so as to train the third preset model. The specific training and learning process is a prior art and will not be repeated here.

When the third preset model is used for prediction, the first prediction result and the flow difference characteristic sequence can be input into the third preset model, and the flow prediction result of the area to be predicted can be obtained by the third preset model. The flow prediction result of the area to be predicted is the flow value corresponding to each prediction time point in the area to be predicted within the preset prediction time.

In this embodiment, the initial flow prediction sequence of the area to be predicted can be obtained through the first preset model, and then the flow difference feature sequence corresponding to each population flow type can be obtained by predicting the initial flow prediction sequence. Finally, the initial flow prediction sequence is corrected using the flow difference feature sequence through the third preset model to obtain the flow prediction result of the area to be predicted, so that the prediction result takes into account multiple characteristics of the area to be predicted and the influence of other areas within the target range, thereby improving the accuracy of the prediction result.

Furthermore, the first prediction result also includes: at least one of: a holiday sequence, a trend sequence and a season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction period, and the season sequence is used to indicate the seasonal characteristics of the area to be predicted within the preset prediction period.

In one embodiment, the initial traffic prediction sequence {Y'} _L of the area to be predicted, as well as at least one of the holiday sequence {H} _L , the trend sequence {T} _L and the season sequence {S} _L can be obtained through the pre-trained prophet model, wherein the holiday sequence {H} _L is used to indicate the holiday characteristics of the area to be predicted within the preset prediction time, the trend sequence {T} _L is used to indicate the trend characteristics of the area to be predicted within the prediction time, and the season sequence {S} _L is used to indicate the seasonal characteristics of the area to be predicted within the preset prediction time. Among them, {Y'} _L , {H} _L , {T} _L and {S} _L are one-dimensional time series of L time points respectively, and the L time points here can be understood as multiple time points within the preset prediction time. For example, assuming that the traffic of the area to be predicted needs to be predicted for the next week, the time granularity is hours, then L is 7*24=168. In this way, when the initial traffic prediction sequence is modified in the third preset model, the influence of the time, holidays and activities of the area to be predicted on the traffic value of the area to be predicted can be considered, so that the traffic prediction result of the area to be predicted is more accurate.

Further, the second preset model includes N sub-models, the second prediction result includes N second prediction sub-results, and N is a positive integer;

The above step 202, inputting the initial traffic prediction sequence and the feature matrix into the second preset model to obtain the second prediction result, includes:

Input the initial traffic prediction sequence and feature matrix into the target sub-model to obtain the second prediction sub-result;

Among them, the target sub-model is any sub-module among the N sub-models.

In one embodiment, the second preset model may include 1, 2, 3, etc. sub-models, and each sub-model corresponds to a second prediction sub-result. In this way, when the initial flow prediction sequence and the feature matrix are input into the target sub-model, a second prediction sub-result can be obtained through the target sub-model, thereby obtaining N second prediction sub-results. The target sub-model here is any sub-module among the N sub-models. For example, assuming that the number of sub-models in the second preset model is 3, which are used to predict the flow difference feature sequence corresponding to the population inflow type, the flow difference feature sequence corresponding to the population outflow type, and the flow difference feature sequence corresponding to the population stability type, then it is only necessary to input the initial flow prediction sequence and the feature matrix of the area to be predicted into these 3 sub-models to obtain the flow difference feature sequences corresponding to the population inflow type, the population outflow type, and the population stability type.

In this embodiment, different sub-models are set to predict the traffic difference characteristic sequences of different population flow types, thereby obtaining the change of network traffic in the predicted area under the influence of areas with different population flow types, so that the predicted traffic value of the predicted area is more accurate.

Furthermore, before the initial traffic prediction sequence and the feature matrix are input into the second preset model to obtain the second prediction result, the method includes:

Train and learn historical traffic data and feature matrices to obtain N sub-models;

Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow, outflow and stable types.

In one embodiment, the population flow type may include at least one of an inflow type, an outflow type, and a stable type, and the present application does not make specific restrictions. For example, when the population flow types of each area within the target range may include three types: an inflow type, an outflow type, and a stable type, the second preset model needs to include three sub-models. Before using these three sub-models for prediction, it is necessary to train and learn the historical flow data and the feature matrix to obtain these three sub-models. Specifically, the historical flow data of each area in the target range is classified according to these three types of population flow, and the flow means corresponding to multiple areas with the same population flow type within the target range at each collection time point are calculated respectively to obtain three flow mean sequences, and then the first flow data of the area to be predicted is subtracted from the three flow mean sequences to obtain three difference feature sequences, and then the historical flow data and the feature matrix of the area to be predicted are used as the input of the three sub-models, and the three difference feature sequences are used as the output of the three sub-models respectively, and three LSTM models are obtained by training.

Of course, when each area within the target range includes two types of population mobility or one type of population mobility, the above method can be used to train to obtain two sub-models or one sub-model in the second preset model.

In this embodiment, multiple sub-models can be set according to the population flow type, and the second multiple sub-models can be trained and learned to achieve the prediction of the impact of multiple areas corresponding to different population flow types on the flow of the predicted area.

Further, referring to FIG3 , FIG3 is a flowchart of the third method for predicting network traffic provided by an embodiment of the present invention. The above steps train and learn the historical traffic data and the feature matrix to obtain N sub-models, including:

Step 301: Classify the historical traffic data of each area according to the population flow type of each area within the target range.

In one embodiment, the population flow types of each region within the target range may include inflow type, outflow type and stable type. The historical flow data of each region may be classified and merged according to the three population flow types to obtain the historical time series corresponding to the three population flow types. Assuming that the historical flow data of each region is a time series of four-dimensional data corresponding to w time collection points, the inflow type time series {W2} _w×4 , the outflow type time series {W3} _w×4 , and the stable type time series {W4} _w×4 can be obtained.

Step 302: calculate N mean sequences corresponding to N population flow types respectively, where the mean sequences are used to indicate the flow mean values corresponding to multiple areas with the same population flow type within the target range at each collection time point.

Calculate the average of the flow values at each collection time point in the inflow time series {W2} _w×4 , the outflow time series {W3} _w×4 , and the stable time series {W4} _w×4 respectively, and obtain the flow mean corresponding to each collection time point, and then obtain the mean sequence {C1} _{w ×2 of the inflow time series {W2} w×4} , the mean sequence {C2} _{w ×2} of the outflow time series {W3} w×4, and the mean sequence {C3} w _×2 of the stable time series {W4} _w×4 , where each mean sequence is a two-dimensional time series including w collection times and the flow mean corresponding to the w collection times.

Step 303: Based on the first flow data and N mean sequences, determine N difference feature sequences corresponding to N population flow types, wherein the difference feature sequences include the difference between the flow value corresponding to each collection time point in the first flow data and the flow mean at the corresponding time point of the mean sequence.

By subtracting the flow value corresponding to each collection time point in the first flow data {W1} _w×4 from the flow mean corresponding to each collection time point in the mean sequence {C1} _w×2 , {C2} _w×2 , {C3} _w×2 , we can obtain the difference feature sequence {D1} _w×2 corresponding to the inflow type, the difference feature sequence {D2} _w×2 corresponding to the outflow type, and the difference feature sequence {D3} _w×2 corresponding to the stable type.

Step 304: Use the first traffic data and the feature matrix as inputs of each of the N sub-models, use the N differential feature sequences as outputs of each of the N sub-models, and obtain N sub-models through training and learning.

Since the population flow types of each region within the target range include inflow type, outflow type and stable type, it is necessary to train the neural network models corresponding to the three population flow types. When training the three neural network models, the first flow data {W1} _w×4 and the feature matrix {F} _w×k can be used as the input of the three neural network models, and the difference feature sequence {D1} _w×2 corresponding to the inflow type, the difference feature sequence {D2} _w×2 corresponding to the outflow type and the difference feature sequence {D3} _w×2 corresponding to the stable type can be used as the output of the three neural network models respectively. The parameters in the three neural network models are trained and learned, so that the three neural network models can be obtained.

Thus, after training to obtain three neural network models, the initial flow prediction sequence {Y'} _L and the feature matrix {F} _w×k can be input into the three neural network models to predict three flow difference feature sequences, namely, the flow difference feature sequence {D1′} _L corresponding to the inflow type, the flow difference feature sequence {D2′} _L corresponding to the outflow type, and the flow difference feature sequence {D3′} _L corresponding to the stable type. In this embodiment, the flow difference feature sequence corresponding to each population flow type can be determined according to the population flow type of each area within the target range, thereby facilitating the subsequent correction of the initial flow prediction sequence of the area to be predicted by the flow difference feature sequence corresponding to different population flow types, so that the flow prediction result of the area to be predicted is more accurate.

Furthermore, the above step 102, based on the first traffic data, determines the feature matrix corresponding to the area to be predicted, including:

Performing feature marking on the first flow data to obtain multiple feature sequences;

Based on multiple feature sequences, determine the feature matrix corresponding to the area to be predicted;

Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values corresponding to the multiple collection time points, holiday information and activity information, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences.

In one embodiment, the first traffic data includes the collection time, the traffic value corresponding to each collection time, the holiday information corresponding to each collection time, and the activity information corresponding to each collection time, which is represented by {W1} _w×4 , where W1 represents the four-dimensional data matrix corresponding to the w time collection points in the area to be predicted, and the number of w is determined by the historical preset time and the time granularity collected. In this way, the first traffic data can be marked with features through big data analysis, and its features include time features, holiday features, activity features and regional features, thereby obtaining a time feature sequence, a holiday feature sequence, an activity feature sequence and a regional feature sequence, and then forming a feature matrix {F} _w×k according to the time feature sequence, the holiday feature sequence, the activity feature sequence and the regional feature sequence, where F represents the k-dimensional data matrix corresponding to the w time collection points, the number of w is determined by the historical preset time and the time granularity collected, and k is an arbitrary positive integer. Specifically, the time feature is used to indicate the time information of each collection time of the first flow data, such as the corresponding season (i.e., spring, summer, autumn or winter), the corresponding time period (i.e., morning, afternoon, noon or evening), whether it is a rush hour, etc.; the holiday feature is used to indicate the holiday information of each collection time of the first flow data, such as whether it is a holiday, the name of the holiday, the duration of the holiday, etc.; the activity feature is used to indicate the activity information of each collection time of the first flow data, such as whether it is an activity day, the name of the activity, the duration of the activity, the scope of the activity area, etc.; the regional feature is used to indicate the information of the area to be predicted, such as the type of population flow (i.e., population inflow type, population outflow type or population stability type), the type of city (i.e., tourist city, agricultural city or industrial city), the gross domestic product (GDP) value of the month, the number of permanent residents in the region, the number of urban population, the number of colleges and universities, the number of primary and secondary schools, etc.

It should be noted that the acquisition of training data for model prediction data can be implemented in the same manner as described above, see FIG. 4, which is a schematic diagram of training data provided by an embodiment of the present invention. As shown in FIG. 4, historical traffic data is first collected, where the historical traffic data represents the historical traffic data of each area in the target range, and the number of collection time points and the granularity of the collection time can be set according to actual needs. Among them, the historical traffic data is time series data including multiple dimensions such as the collection time, the traffic value corresponding to each collection time, the holiday information corresponding to each collection time, and the activity information corresponding to each collection time. After collecting the first traffic data of the area to be predicted, a feature sequence is constructed, wherein the feature sequence may include a time feature sequence, a holiday feature sequence, an activity feature sequence, and a regional feature sequence of the area to be predicted. After the feature sequence is constructed, the historical traffic data of each area, as well as the time feature sequence, the holiday feature sequence, the activity feature sequence, and the regional feature sequence are used as training data for training and learning the preset model.

In this embodiment, the feature matrix of the area to be predicted is constructed by feature marking the first flow data of the area to be predicted. Therefore, when training and prediction are performed through a preset model, the influence of factors such as time characteristics, holiday characteristics, activity characteristics and regional characteristics on the flow value of the area to be predicted can be comprehensively considered, thereby improving the accuracy of the prediction results.

In an application example, see FIG5, which is a schematic diagram of the structure of a preset model provided by an embodiment of the present invention. In FIG5, the preset model includes a first preset model, a second preset model and a third preset model. The first preset model is cascaded with the second preset model and the third preset model, and the second preset model is cascaded with the third preset model. The second preset model includes a first submodule, a second submodel and a third submodel. The first submodule, the second submodel and the third submodel correspond to the inflow type, the outflow type and the stable type of population flow types respectively. The first flow data {W1} _w×4 and the feature matrix {F} _w×k of the area to be predicted are used as inputs of the first preset model, and the first preset model outputs the initial flow prediction sequence {Y'} _L , the holiday sequence {H} _L , the trend sequence {T} _L and the season sequence {S} _L. The initial flow prediction sequence {Y'} _L and the feature matrix {F} _w×k are input to the first submodule, the second submodel and the third submodel, and three flow difference feature sequences are predicted, namely, the flow difference feature sequence {D1′} _L corresponding to the inflow type, the flow difference feature sequence {D2′} _L corresponding to the outflow type and the flow difference feature sequence {D3′} _L corresponding to the stable type. Finally, the initial flow prediction sequence {Y'} _L , the holiday sequence {H} _L , the trend sequence {T} _L , the seasonal sequence {S} _L , and the flow difference feature sequence {D1′} _L corresponding to the inflow type, the flow difference feature sequence {D2′} _L corresponding to the outflow type and the flow difference feature sequence {D3′} _L corresponding to the stable type are used as the input of the third preset model, and the flow prediction result {Y} _L of the area to be predicted is obtained through the third preset model.

In this embodiment, the data such as the initial traffic prediction sequence, holiday sequence, trend sequence and season sequence of the area to be predicted can be obtained through the first preset model, and the traffic difference feature sequence corresponding to each population flow type can be obtained through the second preset model. Finally, the initial traffic prediction sequence is corrected by the third preset model, so that the prediction result takes into account multiple characteristics of the area to be predicted and the impact of population flow in other areas within the target range on the traffic, thereby improving the accuracy of the prediction result. The embodiment of the present invention also provides a network traffic prediction device, see Figure 6, Figure 6 is one of the structural schematic diagrams of the network traffic prediction device provided by the embodiment of the present invention, as shown in Figure 6, the network traffic prediction device 600 includes:

An acquisition module 601 is used to acquire historical traffic data of each area within a target range, where the historical traffic data includes first traffic data corresponding to the area to be predicted;

A first determination module 602 is used to determine a feature matrix corresponding to the area to be predicted based on the first traffic data;

The second determination module 603 is used to input the historical traffic data and the feature matrix into a preset model to determine the traffic prediction result of the area to be predicted;

The preset models include a first preset model, a second preset model and a third preset model. The first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model.

Optionally, the second determining module 603 includes:

A first input submodule, used for inputting the first flow data and the feature matrix into a first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence;

A second input submodule, used to input the initial traffic prediction sequence and the feature matrix into a second preset model to obtain a second prediction result, wherein the second prediction result includes a traffic difference feature sequence;

A third input submodule is used to input the first prediction result and the flow difference feature sequence into a third preset model to obtain a flow prediction result of the area to be predicted, and the flow difference feature sequence is used to indicate the impact of population flow in each area within the target range on the flow of the area to be predicted;

Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model. The third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence.

Optionally, the first prediction result further includes:

At least one of the holiday sequence, trend sequence and season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction period, and the season sequence is used to indicate the seasonal characteristics of the area to be predicted within the preset prediction period.

Optionally, the second preset model includes N sub-models, the second prediction result includes N second prediction sub-results, and N is a positive integer; the second input submodule includes:

An input unit, used to input the initial traffic prediction sequence and the feature matrix into the target sub-model to obtain a second prediction sub-result;

Among them, the target sub-model is any sub-module among the N sub-models.

Optionally, the second determining module 603 further includes:

The training and learning submodule is used to train and learn historical traffic data and feature matrices to obtain N submodels;

Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow, outflow and stable types.

Optionally, the training learning submodule is specifically used for:

Classify the historical flow data of each area according to the population flow type of each area within the target range;

Calculate N mean sequences corresponding to N population flow types respectively. The mean sequence is used to indicate the mean flow rate corresponding to multiple areas with the same population flow type within the target range at each collection time point;

Based on the first flow data and the N mean sequences, determine N difference feature sequences corresponding to the N population flow types, the difference feature sequences including the difference between the flow value corresponding to each collection time point in the first flow data and the flow mean at the corresponding time point of the mean sequence;

The first flow data and the feature matrix are used as inputs of each of the N sub-models, and the N differential feature sequences are used as outputs of each of the N sub-models, and N sub-models are obtained through training and learning.

Optionally, the first determining module 602 includes:

A marking submodule, used for performing feature marking on the first flow data to obtain a plurality of feature sequences;

The second determination submodule is used to determine the feature matrix corresponding to the area to be predicted based on multiple feature sequences;

Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values corresponding to the multiple collection time points, holiday information and activity information, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences.

The network traffic prediction device 600 can implement each process of the above-mentioned network traffic prediction method embodiment and achieve the same beneficial effects. To avoid repetition, it will not be described here.

An embodiment of the present invention also provides a network traffic prediction device 600, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, each process of the above-mentioned network traffic prediction method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

Specifically, as shown in FIG. 7 , an embodiment of the present invention further provides a network traffic prediction device, including a bus 701 , a transceiver 702 , an antenna 703 , a bus interface 704 , a processor 705 and a memory 706 .

Processor 705, used to obtain historical traffic data of each area within the target range, where the historical traffic data includes first traffic data corresponding to the area to be predicted;

Based on the first traffic data, determining a feature matrix corresponding to the area to be predicted;

Input historical traffic data and feature matrix into the preset model to determine the traffic prediction result of the area to be predicted;

The preset models include a first preset model, a second preset model and a third preset model. The first preset model is cascaded with the second preset model and the third preset model respectively, and the second preset model is cascaded with the third preset model.

Furthermore, the processor 705 is further configured to input the first flow data and the feature matrix into a first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence;

Inputting the initial flow prediction sequence and the feature matrix into a second preset model to obtain a second prediction result, wherein the second prediction result includes a flow difference feature sequence;

The first prediction result and the flow difference characteristic sequence are input into the third preset model to obtain the flow prediction result of the area to be predicted, and the flow difference characteristic sequence is used to indicate the influence of the population flow in each area within the target range on the flow of the area to be predicted;

Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model. The third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence.

Furthermore, the first prediction result also includes:

At least one of the holiday sequence, trend sequence and season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction period, and the season sequence is used to indicate the seasonal characteristics of the area to be predicted within the preset prediction period.

Furthermore, the processor 705 is further configured to input the initial traffic prediction sequence and the feature matrix into the target sub-model to obtain a second prediction sub-result;

Among them, the target sub-model is any sub-module among the N sub-models.

Furthermore, the processor 705 is also used to train and learn the historical traffic data and the feature matrix to obtain N sub-models;

Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow, outflow and stable types.

Furthermore, the processor 705 is further configured to classify the historical flow data of each area according to the population flow type of each area within the target range;

Calculate N mean sequences corresponding to N population flow types respectively. The mean sequence is used to indicate the mean flow rate corresponding to multiple areas with the same population flow type within the target range at each collection time point;

Based on the first flow data and the N mean sequences, determine N difference feature sequences corresponding to the N population flow types, the difference feature sequences including the difference between the flow value corresponding to each collection time point in the first flow data and the flow mean at the corresponding time point of the mean sequence;

The first flow data and the feature matrix are used as inputs of each of the N sub-models, and the N differential feature sequences are used as outputs of each of the N sub-models, and N sub-models are obtained through training and learning.

Furthermore, the processor 705 is further configured to perform feature marking on the first flow data to obtain a plurality of feature sequences;

Based on multiple feature sequences, determine the feature matrix corresponding to the area to be predicted;

Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values corresponding to the multiple collection time points, holiday information and activity information, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences.

In FIG. 7 , a bus architecture (represented by bus 701) is shown, and bus 701 may include any number of interconnected buses and bridges, and bus 701 links various circuits including one or more processors represented by processor 705 and memory represented by memory 706. Bus 701 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 704 provides an interface between bus 701 and transceiver 702. Transceiver 702 may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. Data processed by processor 705 is transmitted on a wireless medium via antenna 703, and further, antenna 703 also receives data and transmits the data to processor 705.

The processor 705 is responsible for managing the bus 701 and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 706 can be used to store data used by the processor 705 when performing operations.

Optionally, the processor 705 may be a CPU, an ASIC, an FPGA or a CPLD.

The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above-mentioned network traffic prediction method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims (9)

1. A network traffic prediction method, characterized in that the method comprises: Acquire historical traffic data of each area within the target range, wherein the historical traffic data includes first traffic data corresponding to the area to be predicted; Based on the first traffic data, determining a feature matrix corresponding to the area to be predicted; Inputting the historical traffic data and the characteristic matrix into a preset model to determine the traffic prediction result of the area to be predicted; The preset model includes a first preset model, a second preset model and a third preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, the second preset model is cascaded with the third preset model, and the second preset model is used to predict the flow conversion relationship between the area to be predicted and other areas within the target range; The step of inputting the historical traffic data and the characteristic matrix into a preset model to determine the traffic prediction result of the area to be predicted includes: Inputting the first flow data and the feature matrix into the first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence; Inputting the initial flow prediction sequence and the characteristic matrix into the second preset model to obtain a second prediction result, wherein the second prediction result includes a flow difference characteristic sequence, and the flow difference characteristic sequence is used to indicate the influence of the population flow in each area within the target range on the flow of the area to be predicted; Inputting the first prediction result and the flow difference feature sequence into the third preset model to obtain the flow prediction result of the area to be predicted; Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model, and the third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence. 2. The method according to claim 1, characterized in that the first prediction result further comprises: At least one of the holiday sequence, trend sequence and season sequence of the area to be predicted, wherein the holiday sequence is used to indicate the holiday characteristics of the area to be predicted within a preset prediction time period, the trend sequence is used to indicate the trend characteristics of the area to be predicted within the prediction time period, and the season sequence is used to indicate the season characteristics of the area to be predicted within the preset prediction time period. 3. The method according to claim 1, characterized in that the second preset model includes N sub-models, the second prediction result includes N second prediction sub-results, and N is a positive integer; The step of inputting the initial flow prediction sequence and the feature matrix into a second preset model to obtain a second prediction result includes: Inputting the initial traffic prediction sequence and the feature matrix into the target sub-model to obtain a second prediction sub-result; The target sub-model is any sub-module among the N sub-models. 4. The method according to claim 3, characterized in that before inputting the initial flow prediction sequence and the feature matrix into a second preset model to obtain a second prediction result, it comprises: Performing training and learning on the historical traffic data and the feature matrix to obtain the N sub-models; Among them, the value of N is consistent with the number of population flow types included in each area within the target range, and the population flow types include at least one of inflow type, outflow type and stable type. 5. The method according to claim 4, characterized in that the training and learning of the historical traffic data and the feature matrix to obtain the N sub-models comprises: Classify the historical flow data of each area according to the population flow type of each area within the target range; Calculate N mean sequences corresponding to N population flow types respectively, wherein the mean sequences are used to indicate the flow mean values corresponding to multiple areas with the same population flow type within the target range at each collection time point; Based on the first flow data and the N mean sequences, determine N difference feature sequences corresponding to the N population flow types, the difference feature sequences comprising differences between flow values corresponding to each acquisition time point in the first flow data and flow means at corresponding time points in the mean sequence; The first traffic data and the feature matrix are used as inputs of each of the N sub-models, and the N differential feature sequences are used as outputs of each of the N sub-models, and the N sub-models are obtained through training and learning. 6. The method according to claim 1, characterized in that the step of determining the feature matrix corresponding to the area to be predicted based on the first traffic data comprises: Performing feature marking on the first flow data to obtain a plurality of feature sequences; Based on the multiple feature sequences, determining a feature matrix corresponding to the area to be predicted; Among them, the first traffic data is a four-dimensional time series including multiple collection time points and the traffic values, holiday information and activity information corresponding to the multiple collection time points, and the multiple feature sequences include time feature sequences, holiday feature sequences, activity feature sequences and regional feature sequences. 7. A network traffic prediction device, characterized in that the device comprises: An acquisition module, used to acquire historical traffic data of each area within the target range, wherein the historical traffic data includes first traffic data corresponding to the area to be predicted; A first determination module, configured to determine a feature matrix corresponding to the area to be predicted based on the first traffic data; A second determination module is used to input the historical traffic data and the characteristic matrix into a preset model to determine the traffic prediction result of the area to be predicted; The preset model includes a first preset model, a second preset model and a third preset model, the first preset model is cascaded with the second preset model and the third preset model respectively, the second preset model is cascaded with the third preset model, and the second preset model is used to predict the flow conversion relationship between the area to be predicted and other areas within the target range; The second determining module includes: A first input submodule, used for inputting the first flow data and the feature matrix into the first preset model to obtain a first prediction result, wherein the first prediction result includes an initial flow prediction sequence; A second input submodule is used to input the initial traffic prediction sequence and the feature matrix into the second preset model to obtain a second prediction result, wherein the second prediction result includes a traffic difference feature sequence, and the traffic difference feature sequence is used to indicate the impact of population flow in each area within the target range on the traffic of the area to be predicted; A third input submodule, used for inputting the first prediction result and the flow difference feature sequence into the third preset model to obtain the flow prediction result of the area to be predicted; Among them, the first preset model and the second preset model are different machine learning models, and the third preset model is a correction model, and the third preset model is used to correct the initial traffic prediction sequence according to the traffic difference feature sequence. 8. A network traffic prediction device, characterized in that it comprises: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the network traffic prediction method as described in any one of claims 1 to 6 are implemented. 9. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the network traffic prediction method according to any one of claims 1 to 6 are implemented.