Flow prediction method and system

Abstract

The invention provides a traffic prediction method and system, and belongs to the technical field of traffic prediction, and the method comprises the steps: collecting historical traffic data of a to-be-predicted user, carrying out the data processing, and obtaining discrete feature information and continuous feature information before a prediction time period in a preset period; inputting the discrete feature information into a pre-constructed first traffic prediction model for traffic prediction to obtain a first traffic prediction result corresponding to a prediction time period in a preset period; and inputting the continuous feature information and the first traffic prediction result into a pre-constructed second traffic prediction model for traffic prediction to obtain a final traffic prediction result corresponding to the prediction time period. The method has the beneficial effects that the machine learning fusion model is adopted to predict the traffic time sequence data scene, more comprehensive and more accurate prediction of the traffic time sequence data scene is realized, the limitation of a single model in a complex data scene is overcome, and the prediction accuracy and robustness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of flow prediction, and in particular to a flow prediction method and system. Background Art

[0002] With the rapid development of mobile communication technology and the rapid popularization of intelligent mobile devices (such as smartphones), users are using mobile devices more and more frequently, generating a large amount of wireless network traffic. Due to the immediacy and dynamism of data traffic consumption in user communication packages, there are high requirements for the timeliness of insights into user traffic demand. By predicting user traffic usage, early warning of groups that are about to overflow, and providing user traffic reminder services in advance, it also brings considerable benefits to the operation strategy of mobile network data traffic and the sales of data traffic packages.

[0003] At present, time series analysis methods such as the autoregressive integrated moving average (ARIMA) model are mostly used to predict traffic data. The ARIMA model performs well in processing stationary or weakly stationary time series and can effectively capture the linear trend and seasonal changes of the data. However, in the face of massive, high-frequency and volatile traffic data in modern communication networks, the ARIMA model is only suitable for capturing linear relationships. The prediction accuracy for nonlinear, complex and changeable traffic data is limited, and the model training and prediction process are inefficient, making it difficult to meet large-scale and real-time requirements for operational scenarios.

[0004] In order to overcome the shortcomings of traditional traffic prediction methods such as ARIMA, deep learning technology, especially recurrent neural networks (RNN) and their variants such as long short-term memory networks (LSTM), has been widely used in the field of traffic prediction in recent years. LSTM can effectively process dependencies in long sequence data and capture complex nonlinear patterns through its unique gating mechanism, thereby achieving high-precision traffic prediction. However, the training and application of deep learning models have relatively high requirements on the resource configuration and investment cost of the computing platform, such as GPU resources. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a flow prediction method and system.

[0006] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0007] A flow prediction method, comprising:

[0008] Step S1, collecting historical traffic data of the user to be predicted;

[0009] Step S2, processing the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the predicted time period within a preset period;

[0010] Step S3, inputting the discrete characteristic information into a pre-built first traffic prediction model to perform traffic prediction, and obtaining a first traffic prediction result corresponding to a prediction time period within the preset period;

[0011] Step S4, inputting the continuous characteristic information before the prediction time period within the preset period and the first traffic prediction result into a pre-built second traffic prediction model to perform traffic prediction, and obtaining the final traffic prediction result corresponding to the prediction time period.

[0012] Preferably, the historical traffic data includes one or more combinations of user basic information, traffic usage preference behavior information and historical traffic usage information.

[0013] Preferably, step S2 comprises:

[0014] Step S21, preprocessing the historical traffic data to obtain sample data;

[0015] Step S22, performing feature engineering on the sample data, and screening to obtain the discrete feature information and the continuous feature information.

[0016] Preferably, in step S21, the preprocessing includes one or more combinations of data missing processing, outlier processing, and data standardization processing.

[0017] Preferably, the first traffic prediction model is a stacked fusion tree model;

[0018] The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; wherein the random forest model and the extreme gradient boosting model serve as base models of the stacked fusion tree model, and the lightweight gradient boosting model serves as a secondary base model of the stacked fusion tree model.

[0019] Preferably, the second traffic prediction model is a linear regression model.

[0020] Preferably, the discrete feature information includes one or more combinations of game preference tags, video preference tags, free traffic in the current preset period, traffic usage in the current preset period, bill consumption in the past N preset periods, traffic usage in the past N preset periods, whether traffic overflow occurred in the previous preset period, and whether traffic was suppressed in the previous preset period, where N is a positive integer greater than 0.

[0021] Preferably, the continuous characteristic information is flow time series information.

[0022] The present invention also provides a flow prediction system for implementing the flow prediction method as described above, the flow prediction system comprising:

[0023] A data collection module is used to collect historical traffic data of users to be predicted;

[0024] A data processing module, connected to the data collection module, for processing the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the predicted time period within a preset period;

[0025] A first traffic prediction module, connected to the data processing module, for inputting the discrete characteristic information into a pre-built first traffic prediction model to perform traffic prediction, and obtain a first traffic prediction result corresponding to a prediction time period within a preset period;

[0026] The second traffic prediction module is respectively connected to the data processing module and the first traffic prediction module, and is used to input the continuous characteristic information before the prediction time period within the preset period and the first traffic prediction result into a pre-constructed second traffic prediction model for traffic prediction, so as to obtain the final traffic prediction result corresponding to the prediction time period.

[0027] Preferably, the first traffic prediction model is a stacked fusion tree model;

[0028] The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; wherein the random forest model and the extreme gradient boosting model serve as base models of the stacked fusion tree model, and the lightweight gradient boosting model serves as a secondary base model of the stacked fusion tree model;

[0029] The second traffic prediction model is a linear regression model.

[0030] The advantages or beneficial effects of the technical solution of the present invention are:

[0031] The present invention adopts a machine learning fusion model to predict traffic time series data scenarios. The first traffic prediction model makes preliminary predictions based on discrete feature information to quickly capture the general trend of user traffic usage; the second traffic prediction model further utilizes continuous feature information and preliminary prediction results for deep mining and correction to achieve a more comprehensive and accurate prediction of traffic time series data scenarios, overcoming the limitations of a single model in complex data scenarios and significantly improving the accuracy and robustness of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of a flow prediction method in a preferred embodiment of the present invention;

[0033] Figure 2 A schematic diagram of a data processing flow in a preferred embodiment of the present invention;

[0034] Figure 3 A schematic diagram of the architecture of traffic prediction in a preferred embodiment of the present invention;

[0035] Figure 4 A schematic diagram of base model training in the first traffic prediction model in a preferred embodiment of the present invention;

[0036] Figure 5 A schematic diagram of the training of the secondary basis model in the first traffic prediction model in a preferred embodiment of the present invention;

[0037] Figure 6 This is a structural block diagram of a traffic prediction system in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0038] 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 only 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.

[0039] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0040] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0041] Glossary:

[0042] ARIMA (Autoregressive Integrated Moving Average Model): is a time series forecasting method. The so-called ARIMA model refers to a model that converts a non-stationary time series into a stationary time series, and then regresses the dependent variable only on its lagged value and the present value and lagged value of the random error term.

[0043] RNN (Recurrent Neural Network): It is an artificial neural network whose nodes are connected in a loop. The internal state of this network can show dynamic temporal behavior. Unlike feedforward neural networks, RNN can use its internal memory to process input sequences of arbitrary temporal order, including but not limited to complex tasks such as non-segmented handwriting recognition and speech recognition.

[0044] LSTM (Long Short-Term Memory): A time recursive neural network suitable for processing and predicting important events with relatively long intervals and delays in time series. The LSTM model effectively solves the gradient vanishing or exploding problem faced by traditional RNNs when processing long-distance dependencies by introducing a "gate" control mechanism (including forget gate, input gate, and output gate).

[0045] Random forest (RF): is a classifier that uses multiple decision trees to train and predict samples. It achieves classification or prediction of samples based on the construction and combination of multiple decision trees. It improves the overall classification or regression performance by training multiple decision trees in parallel and combining their prediction results.

[0046] XGBoost is an optimized distributed gradient boosting library designed to be efficient, flexible, and portable. It implements machine learning algorithms under the Gradient Boosting framework. XGBoost provides parallel tree boosting (also known as GBDT, GBM) that can solve many data science problems quickly and accurately. The same code runs on major distributed environments (Hadoop, SGE, MPI) and can solve problems beyond billions of examples.

[0047] Light GBM (Light Gradient Boosting Machine) is a gradient boosting framework that uses decision trees as base learners.

[0048] See also Figure 1 In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a flow prediction method is provided, comprising:

[0049] Step S1, collecting historical traffic data of the user to be predicted;

[0050] Among them, historical traffic data includes but is not limited to user basic information, traffic usage preference behavior information and historical traffic usage information.

[0051] Basic user information includes but is not limited to user ID, age and gender, device type, and package type. User ID is used to uniquely identify each user; device type, such as smartphones, tablets, or laptops, has different traffic consumption patterns; package type refers to the traffic package subscribed by the user, such as unlimited package and pay-as-you-go package;

[0052] Traffic usage preference behavior information includes but is not limited to access time preferences, application usage preferences, and location preferences. Access time preferences are used to indicate the time period in which users usually use more traffic, such as morning commuting, lunch break, and evening entertainment; application usage preferences are used to indicate the types of applications that users use most frequently, such as social media, video platforms, online games, office software, etc.; location preferences are used to indicate the network coverage of users' permanent residence and frequently visited places, which affects the way traffic is used, such as using Wi-Fi at home and relying on mobile data when going out.

[0053] Historical traffic usage information includes, but is not limited to, historical account periods in the past N preset periods and historical traffic usage values ​​before the predicted time period within the preset period, where N is a positive integer greater than 0. N can be set according to actual needs, such as selecting the past 3 months, 6 months, or 12 months as the observation period. Each cycle usually corresponds to a billing cycle, such as one cycle for each month. For each preset cycle, it is also necessary to collect traffic usage at a certain predicted time point within the cycle or a period of time before the predicted time period (such as the beginning, middle, or end of the month), that is, the traffic that has been generated but not billed within the preset period. For example, to predict the traffic usage in the last 10 days of this month, the collected historical traffic usage information may include the historical account periods of the past 3 months and the traffic usage in the first 20 days of this month.

[0054] Step S2, processing the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the prediction time period within a preset period;

[0055] Among them, data processing includes but is not limited to data preprocessing and feature engineering.

[0056] Data preprocessing includes but is not limited to data missing processing, outlier processing, and data standardization. In the missing data processing step, by traversing all fields in the data set, the number of missing values ​​in each field is recorded. Depending on the specific situation of the data, you can choose to fill (such as using the mean, median, mode, or filling according to business logic), or delete records with missing values ​​(if the proportion of missing values ​​is not high). In the outlier processing step, use statistical methods (such as Z-score, IQR, etc.) or rules based on business understanding to identify outliers, and clean up outliers (such as deleting, replacing with mean / median, etc.). In the data standardization processing step, for continuous data, such as traffic usage, use z-score standardization or min-max normalization to eliminate the impact of different dimensions on model training.

[0057] Feature engineering includes but is not limited to feature engineering of preprocessed sample data, and screening several features with high importance coefficients through feature selection, such as discrete feature information and continuous feature information, as model input.

[0058] Discrete feature information includes but is not limited to one or more combinations of game preference tags, video preference tags, free traffic in the current preset period, traffic usage in the current preset period, bill consumption in the past N preset periods, traffic usage in the past N preset periods, whether traffic overflow occurred in the previous preset period, and whether traffic was suppressed in the previous preset period.

[0059] The continuous feature information is the traffic time series information. As an example but not a limitation, it can refer to the traffic usage in the first 20 days of this month as shown above.

[0060] Step S3, inputting the discrete feature information into a pre-built first traffic prediction model to perform traffic prediction, and obtaining a first traffic prediction result corresponding to a prediction time period within a preset period;

[0061] Step S4, inputting the continuous characteristic information before the prediction time period within the preset cycle and the first flow prediction result into a pre-built second flow prediction model to perform flow prediction, and obtaining the final flow prediction result corresponding to the prediction time period.

[0062] Specifically, the traffic prediction of the present invention is carried out in two stages, namely, the first traffic prediction model and the second traffic prediction model. First, the screened and processed discrete feature information, such as game preference tags, video preference tags, free traffic volume, usage volume, bill consumption, whether traffic overflow occurs, etc. in the current preset period, is used as the input of the first traffic prediction model to output the first traffic prediction result of the prediction time period in the preset period, which is a predicted value, such as the traffic usage in the last 10 days of this month.

[0063] Then, the continuous feature information before the prediction time period within the preset period (such as the traffic usage in the first 20 days of this month) and the output of the first traffic prediction model (such as the predicted traffic usage in the last 10 days of this month) are used as the input of the second traffic prediction model, and the model output is used as the final traffic prediction result.

[0064] The present invention uses a machine learning fusion model to predict traffic time series data scenarios through insights into user traffic usage patterns and other communication portraits, overcoming the limitations of a single model in complex data scenarios and significantly improving the accuracy and robustness of the prediction.

[0065] As a preferred embodiment, wherein Figure 2 As shown, step S2 includes:

[0066] Step S21, preprocessing the historical traffic data to obtain sample data;

[0067] Step S22, perform feature engineering on the sample data, and screen to obtain discrete feature information and continuous feature information.

[0068] As a preferred implementation, in step S21, the preprocessing includes one or more combinations of data missing processing, outlier processing, and data standardization processing.

[0069] As a preferred implementation, the discrete feature information includes one or more combinations of game preference tags, video preference tags, free traffic in the current preset period, traffic usage in the current preset period, bill consumption in the past N preset periods, traffic usage in the past N preset periods, whether traffic overflow occurred in the previous preset period, and whether traffic was suppressed in the previous preset period, where N is a positive integer greater than 0.

[0070] As a preferred implementation, the continuous characteristic information is flow time series information.

[0071] As a preferred embodiment, wherein Figure 3 , Figure 4 and Figure 5 As shown, the first traffic prediction model is a stacked fusion tree model;

[0072] The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; among them, the random forest model and the extreme gradient boosting model are used as base models of the stacked fusion tree model, and the lightweight gradient boosting model is used as a secondary base model of the stacked fusion tree model.

[0073] As a preferred implementation, the second traffic prediction model is a linear regression model.

[0074] Specifically, the traffic prediction model of the embodiment of the present invention is composed of two layers, namely: a primary model (ie, the first traffic prediction model mentioned above) and a secondary model (ie, the second traffic prediction model mentioned above).

[0075] In the primary model, the stacking fusion method is used to fuse the three tree model algorithms of random forest, extreme gradient boosting model (XGBoost), and lightweight gradient boosting model (LightGBM), and then the discrete feature information after screening is put into the stacking fusion tree model for model training to obtain the first traffic prediction result. Specifically, the random forest algorithm and XGBoost algorithm are used as the base model training, and LightGBM is used as the secondary base model training. Figure 4 and Figure 5 shown. Figure 4 In the figure, the predictions of A1 are stacked vertically for the 5 points marked on the left, and B1 takes the average of the 5 columns on the left.

[0076] In the secondary model, the prediction results output by the primary model and the continuous feature information are input into the linear regression model for training to obtain the final prediction results.

[0077] Patents for traffic prediction in current communication packages mostly use time series sliding windows (such as ARIMA), complex deep learning algorithms (LSTM), and deep learning algorithm model fusion. ARIMA can only capture linear relationships and learn data sets with small fluctuations in the original data, and the model prediction error is large; LSTM can model complex multivariate sequences, with high operating efficiency and small prediction errors, but it consumes GPU resources. The random forest, LightGBM, and XGBoost algorithms used in the embodiments of the present invention do not require GPU resources, and are suitable for classification and regression algorithms for various data sets. They have good processing capabilities for high-dimensional data sets and excellent generalization performance; at the same time, the algorithm principles of different machine learning models are not the same. Random forest improves performance by reducing model variance. XGBoost is an improvement on the Boosting algorithm based on GBDT, and adds a regular term of model complexity to improve model performance. LightGBM improves performance by reducing model deviation; the Stacking integration algorithm combines the advantages of individual learners to improve the generalization ability of the model and further optimize the overall model performance.

[0078] See also Figure 6 The present invention also provides a flow prediction system for implementing the flow prediction method as described above, the flow prediction system comprising:

[0079] Data collection module 1, used to collect historical traffic data of users to be predicted;

[0080] The data processing module 2 is connected to the data collection module 1 and is used to process the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the prediction time period within a preset period;

[0081] The first traffic prediction module 3 is connected to the data processing module 2 and is used to input the discrete characteristic information into a pre-built first traffic prediction model to perform traffic prediction and obtain a first traffic prediction result corresponding to a prediction time period within a preset period;

[0082] The second traffic prediction module 4 is respectively connected to the data processing module 2 and the first traffic prediction module 3, and is used to input the continuous characteristic information before the prediction time period within the preset period and the first traffic prediction result into the pre-constructed second traffic prediction model for traffic prediction, and obtain the final traffic prediction result corresponding to the prediction time period.

[0083] As a preferred implementation, the first traffic prediction model is a stacked fusion tree model;

[0084] The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; wherein the random forest model and the extreme gradient boosting model are used as base models of the stacked fusion tree model, and the lightweight gradient boosting model is used as a secondary base model of the stacked fusion tree model;

[0085] The second traffic prediction model is a linear regression model.

[0086] Specifically, the embodiment of the present invention aims to provide a user mobile traffic usage prediction model based on machine learning algorithms and Stacking fusion integration models for users who do not have GPU resource capabilities, and can integrate multiple characteristics such as traffic usage and application preferences. The traffic prediction model adopts a two-layer cascade model. The first-level model uses the Stacking method to integrate the random forest, XGBoost and LightGBM algorithms. The second-level model uses a linear regression model, which can support the input of more discrete features and save the GPU overhead of calling conventional deep algorithms.

[0087] In the above preferred embodiment, the present invention mines the user's traffic usage preferences and historical traffic usage behaviors, and uses a machine learning integrated algorithm to learn and predict the user's traffic usage for the current month. The implementation steps are as follows:

[0088] First, the original data required for modeling is collected through the data acquisition platform.

[0089] Next, explore the raw data:

[0090] Overall view: data columns and rows, number of features;

[0091] Data content; data default values ​​and abnormal values;

[0092] Single feature view: data type and data distribution of the feature;

[0093] Correlation of multiple features: Several related features are jointly explored in the image to check the distribution relationship.

[0094] Next, data preprocessing:

[0095] Abnormal data processing: remove unique attributes and handle missing values ​​through three methods, such as directly using features with missing values, deleting features with missing values, and completing missing values;

[0096] Outlier identification and processing: Use the 3σ principle to identify outliers, and there are two ways to deal with outliers: deletion and modification;

[0097] Feature encoding: One-Hot Encoding is used to encode text data.

[0098] Next, feature selection: select relevant feature subsets from a given feature set to reduce the feature vector dimension and alleviate the dimensionality curse problem, thereby reducing the difficulty of the learning task and improving model performance. However, it is necessary to ensure that important features are not lost.

[0099] Then, data modeling: First, use the Stacking fusion method to fuse the three tree model algorithms of random forest, XGBoost, and LightGBM, then put the filtered classification features into the fusion model for model training, and get the prediction results of the first-level model. Finally, put the prediction results of the first-level model and the remaining continuous features into the linear regression model for training to get the final prediction results.

[0100] Finally, forecast label operation and use: upload the final forecast results to the label platform so that frontline personnel can use the forecast data conveniently and quickly. At the same time, establish a regular update mechanism to maintain the forecast results with monthly regular updates.

[0101] By using the system design of the present invention, the hardware cost of deploying large servers and GPUs can be reduced, while the level of care for mobile user traffic can be improved, thereby increasing the sales of mobile traffic packages.

[0102] Experiments have shown that, for the same batch of data, the model evaluation result using the modeling method of the present invention is 55% higher than that using only the linear regression algorithm to model.

[0103] The advantages or beneficial effects of adopting the above technical solution are: the present invention adopts a machine learning fusion model to predict traffic time series data scenarios. The first traffic prediction model makes a preliminary prediction based on discrete feature information to quickly capture the general trend of user traffic usage; the second traffic prediction model further utilizes continuous feature information and preliminary prediction results to conduct in-depth mining and correction to achieve a more comprehensive and accurate prediction of traffic time series data scenarios, overcoming the limitations of a single model in complex data scenarios and significantly improving the accuracy and robustness of the prediction.

[0104] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. A flow prediction method, characterized in that: include: Step S1, collecting historical traffic data of the user to be predicted; Step S2, processing the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the predicted time period within a preset period; Step S3, inputting the discrete characteristic information into a pre-built first traffic prediction model to perform traffic prediction, and obtaining a first traffic prediction result corresponding to a prediction time period within the preset period; Step S4, inputting the continuous characteristic information before the prediction time period within the preset period and the first traffic prediction result into a pre-built second traffic prediction model to perform traffic prediction, and obtaining the final traffic prediction result corresponding to the prediction time period.

2. The flow prediction method according to claim 1, characterized in that: The historical traffic data includes one or more combinations of user basic information, traffic usage preference behavior information, and historical traffic usage information.

3. The flow prediction method according to claim 1, characterized in that: The step S2 comprises: Step S21, preprocessing the historical traffic data to obtain sample data; Step S22, performing feature engineering on the sample data, and screening to obtain the discrete feature information and the continuous feature information.

4. The flow prediction method according to claim 3, characterized in that: In step S21, the preprocessing includes one or more combinations of data missing processing, outlier processing, and data standardization processing.

5. The flow prediction method according to claim 1, characterized in that: The first traffic prediction model is a stacked fusion tree model; The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; wherein the random forest model and the extreme gradient boosting model serve as base models of the stacked fusion tree model, and the lightweight gradient boosting model serves as a secondary base model of the stacked fusion tree model.

6. The flow prediction method according to claim 1, characterized in that: The second traffic prediction model is a linear regression model.

7. The flow prediction method according to claim 1, characterized in that: The discrete feature information includes one or more combinations of game preference tags, video preference tags, free traffic in the current preset period, traffic usage in the current preset period, bill consumption in the past N preset periods, traffic usage in the past N preset periods, whether traffic overflow occurred in the previous preset period, and whether traffic was suppressed in the previous preset period, where N is a positive integer greater than 0.

8. The flow prediction method according to claim 1, characterized in that: The continuous characteristic information is flow time series information.

9. A traffic prediction system, characterized in that: Used to implement the traffic prediction method according to any one of claims 1 to 9, the traffic prediction system comprises: A data collection module is used to collect historical traffic data of users to be predicted; A data processing module, connected to the data collection module, for processing the historical traffic data to obtain discrete characteristic information and continuous characteristic information before the predicted time period within a preset period; A first traffic prediction module, connected to the data processing module, for inputting the discrete characteristic information into a pre-built first traffic prediction model to perform traffic prediction, and obtain a first traffic prediction result corresponding to a prediction time period within a preset period; The second traffic prediction module is respectively connected to the data processing module and the first traffic prediction module, and is used to input the continuous characteristic information before the prediction time period within the preset period and the first traffic prediction result into a pre-constructed second traffic prediction model for traffic prediction, so as to obtain the final traffic prediction result corresponding to the prediction time period.

10. The traffic prediction system according to claim 9, characterized in that: The first traffic prediction model is a stacked fusion tree model; The stacked fusion tree model includes a random forest model, an extreme gradient boosting model and a lightweight gradient boosting model; wherein the random forest model and the extreme gradient boosting model serve as base models of the stacked fusion tree model, and the lightweight gradient boosting model serves as a secondary base model of the stacked fusion tree model; The second traffic prediction model is a linear regression model.

Images