Analysis of Weather Lagged Effects on Freeways Free-Flow Characteristics in Jilin

摘要

ABSTRACTWeather conditions have considerable impact on freeway free-flow characteristics, severalempirical studies have stated that precipitation, snow, and visibility loss may cause reductions inspeed and capacity. Therefore, identifying promising predictors among these meteorologicalfactors is a crucial issue for traffic flow prediction. However, the complex features (irregularities,volatility, trends and noise) inherent in temporally aggregated observed data makes it a ratherdifficult problem. Moreover, in contrast to the development of predictive models to determinethe precise numerical information of traffic parameters, it may be more meaningful to predict thefree-flow trend on major freeways with reasonable accuracy. Besides, in the particular case oftime series forecasting, another crucial element necessary to determine is lagged effects ofweather conditions on free-flow. Therefore, a detailed investigation in this paper was carried outto examine the linkages between meteorological factors and key traffic stream parameters. Thestudy was based on recent archived data from sensor devices, such as inductive loop detectorsand weather sensors, located on provincial freeways of Jilin Province in China. The trend andcyclical components were firstly separated from weather and free-flow parameter series by usingfiltering technique. Then a multiple-equation system known as a vector autoregression (VAR)was proposed for characterizing the temporal dynamics inherent in these components, whileGranger causality theory was adopted to identify the existence of a systemic causal relationshipfor attribute selection. Furthermore, the recently developed method of impulse response functionprovided insight into the lagged effects of these traffic parameters and their responses to weatherconditions, and the multiple series data are reconstructed by incorporating the lagged periods.Finally, various classification models are compared in terms of trend prediction accuracy,Receiver Operating Characteristic (ROC) curve. As a result, K-NN model outperforms the othersfor the overall predication performance, while results also indicate a significant performanceimprovement for these models by incorporating the lagged effects. Besides, some interestingresults were also concluded from our study, including descriptions of the dynamic interplayamong variables, as well as the possible variations in hourly freeway traffic activities withrespect of weather trends. It is hoped that this study will shed light on a fully understanding ofhow weather factors affect freeway traffic conditions.