Analysis And Forecast Of Urban Macro Traffic State Based On ARIMA-SVM Mixed Model

This article analyzes the influencing factors of urban traffic status from a macro level and predicts the future traffic status.The research object is the Beijing Urban Traffic Index.By analyzing and predicting the traffic index and digging up its internal laws,it can provide a scientific reference for the future macro-traffic state prediction and the formulation of traffic policies.At present,there are few studies on macroscopic traffic conditions at home and abroad,and there are only a few studies on traffic indexes.The research content of this article includes the following three aspects:(1)According to the time series chart of Beijing Urban Transportation Index from April 2015 to April 2016,analyze its influencing factors.The changes in the traffic index are affected by factors such as weekends,tail number limit,single and double number limit,long vacation,temperature,precipitation,air quality index,and vehicle ownership,and were analyzed for correlation and significance respectively,and found different The influencing factors have different influences on the traffic index.Among them,the most influential factors are tail number limit line,single and double number limit line and long holiday.(2)Later,this article uses a variety of classic time series models to predict the Beijing urban traffic index,including exponential smoothing,ARIMA model,multiple linear regression model,and dynamic regression model.After analyzing the shortcomings of the seasonal ARIMA model,the ARIMA-SVM hybrid model is proposed.The optimized model uses SVM support vector machine to predict the residual of the ARIMA model,and combines the residual predicted value with the predicted value of the ARIMA model to obtain the final result.After analyzing the multiple linear regression model and the dynamic regression model,the Time series decomposition dynamic regression model.Considering that the traffic index time series has seasonality,and the dynamic regression model is not suitable for seasonal data,the time series is decomposed on the original series,and the remaining parts are predicted by the dynamic regression model after the seasonal parts are removed.Join the season section to get the final result.(3)The predictions of the two optimized combined models for time series are significantly better than the single model.Finally,this paper compares and analyzes the two optimized models.The results show that the dynamic regression model based on time series decomposition has small fluctuations in the sequence More sensitive,but overall,the ARIMA-SVM model is more accurate in predicting the traffic index,with an average absolute error rate of only about 6%.