Integrated Knowledge-Based System for Real-Time Estimation of Incident Durations and Nonrecurrent Congestion Delay for Freeway Networks.

摘要

This study presents a set of models for predicting incident duration and identifying associated variables in the state of Maryland. The incident database from Year 2003 to Year 2005 from Maryland State Highway (MDSHA) Administration was used for model development, and Year 2006 data was used for the model validation. This study has employed the Rule-Based Tree method to develop the primary prediction model. To enhance the prediction accuracy for some types of incidents of complex nature or having limited samples, the study has also calibrated several supplemental components based on the Multinomial Logit and Regression methods. Further exploration for fatality incidents has also been conducted with Naive Bayesian Classifier using the Accident Report database provided by the MD State Police Department. In addition, this study has developed preliminary models for computing incident-induced delay and queue length as a part of applications of incident duration. The developed set of models offers an effective tool for responsible agencies to estimate the approximate predicted duration of a detected incident and its resulting delay as well as queue length.