Heteroscedasticity Analysis And Short Term Forecasting For Continuum Traffic Flow Conditions

Rapid urbanization and vehicle population increase result in serious congestion issue in the surface transportation systems. As one of the effective methods of mitigating this issue, Intelligent Transportation Systems (ITS) developed rapidly over the past decades. Proactive, reliable, and online traffic management and control system is one of the core components of ITS, and its development and construction relies on effective, reliable, and real time short term prediction of traffic condition, including the first order moment, i.e., mean, prediction and the second order moment, i.e., prediction interval, prediction. Over the years, the first order moment prediction has been investigated extensively, while the investigation into the second order moment prediction is limited, remaining in its infancy. In this dissertation, based on the detailed review on the state-of-the-art of the studies on the first order moment investigations, the second order moment modeling and real time prediction of traffic condition series are investigated, with the primary findings as below.First, an extensive literature review is conducted, finding that the time domain time series analysis method is capable of modeling explicitly the first order moment of traffic condition series, and combined with the Kalman filter technique, real time and effective short term mean traffic condition prediction can be achieved. In addition, the primary methods of modeling the second order moment of traffic condition series are identified as Bootstrap, GARCH, and SV, pointing out that GARCH is the most widely applied approach while further studies are still needed in modeling and predicting the second order moment of traffic condition series.Second, heteroscedasity of traffic condition series is investigated systematically, including three aspects of within-group heteroscedasticity analysis, impact of data aggregation time interval on heteroscedasticity of traffic condition series, and seasonal pattern of the second order moment of traffic condition series. It is found that within-group traffic heteroscedasticity series shows pronounced regularity, which can be modeled and predicted using time domain time series model. In addition, traffic condition series exhibits significant conditional heteroscedasticity over a variety of data aggregation time intervals, and with the increase of the time interval, the conditional heteroscedasticity can be reduced to a certain extent while it cannot be fully removed. Moreover, the second order conditional moment has pronounced seasonal pattern, and this pattern should be considered when modeling the second order conditional moment.Third, based on the second order seasonality of traffic condition series, a seasonal adjustment factor approach is proposed. The equations for computing these factors are provided, the characteristics of these factors are analyzed, and the effect of these factors handling the second order seasonality is demonstrated. Through comparing with the conventional GARCH model, the proposed seasonal adjustment factor approach is shown to be able to improve the second order moment modeling performance substantially.Finally, targeting the requirement of real time processing of traffic management and control, an integrated time series model for traffic condition series is proposed, including the first order component and second order component, and bearing primarily the second order moment characteristics in mind and considering the stability of seasonal adjustment factors, a real time prediction method is proposed through applying the adaptive Kalman filter approach. Results based on real world data analysis show that the proposed method can predict in real time the second order characteristics of traffic condition series and construct effectively the prediction interval, validating the proposed online method.In summary, the findings of this dissertation in analyzing and predicting the heteroscedasticity in traffic condition series will facilitate the development and construction of proactive, reliable, and real time traffic management and control systems, which will show significant theoretical meaning and pragmatic value in combating the increasing congestion issues that are prevailing in the cities of China.