A traffic prediction method based on periodic dependence

Abstract

The present invention relates to a method for forecasting traffic volume based on periodic dependence. By collecting the background data of a call center and performing preprocessing to obtain the traffic volume, the periodic dependence of the traffic volume in different time units is used for feature extraction. Add feature dimensions to the extracted features, input all features and corresponding target values ​​into the algorithm model, obtain a stable model after training, input traffic forecast request data into the model, and obtain traffic forecast. The present invention does not depend on the standard sequence model, fully considers the periodic dependence of the traffic volume on different time units, and can well reflect the changing trend of the traffic volume in different time periods; the present invention is based on big data analysis, and Using machine learning algorithms for model training can more accurately predict the traffic volume in the future time period, and can predict the traffic volume in each time period in the long-term in the future.

Description

technical field [0001] The present invention relates to telephone communication; the technical field of automatic or semi-automatic switching offices, in particular to a method for predicting traffic volume based on periodic dependence. Background technique [0002] Traffic forecast refers to a mathematical method to obtain the basic data necessary for medium and long-term development planning of a telecommunication network or to adjust the organization of each trunk circuit group in the near future. Forecast content Traffic forecast can be divided into two parts: total forecast and traffic flow forecast. The total forecast is based on different requirements for a bureau, a city, a province (region), or the country. Traffic flow prediction is to predict the traffic flow from an originating office to a receiving office. It has directionality, so it predicts the distribution of traffic among each office. [0003] Today, more and more companies have established their own call ...

AI Technical Summary

Problems solved by technology

[0006] 1) Time series modeling and forecasting, the future electricity is modeled and predicted in time series; the defect of this method is that it usually focuses on the modeling of the total daily or even monthly electricity, and relies on the standard sequence model, not in the long-term traffic forecast in each time period;

[0007] 2) Poisson modeling prediction, the call process is modeled as a Poisson process, and the call process of a day is modeled as a non-homogeneous Poisson process; the defect is that it is only modeled for the call volume of the next day, without considering the periodicity Dependent factors, although the forecast is the amount of each time period of the day, it is not long-term;

[0008] 3) Linear fixed, mixed modeling forecasting, which introduces the number of incoming calls in the same time period of the previous few time periods and the same time period of the previous few days for parameter estimation; the defect is that the forecast is modeled by day, and periodic dependence factors are not considered;

[0009] 4) Periodic average forecast, taking the average value according to different periods, such as taking the average value according to the daily cycle; the defect is that the periodic average reflects the overall change trend, and cannot accurately predict changes in the future time period

Images