A study on ANN-based real-time accident duration and queue length prediction methodology

Abstract

The increasing of population and vehicles has led to the problem of traffic delay which is one of major issues on the freeways. Also, serious traffic accidents cause severe delays and traffic flow interruption, not only the location in which the accident occurs but area of the surroundings. To be prompt in dealing with the traffic accident and reduce delay time, although it has been utilized the on-line incident detection algorithm, the real-time accident duration and queue length prediction have not been applied. This study reviews the concept of artificial neural network, genetic algorithm, and combination of them to find out optimal solution. The study also reviews the previous studies to find out the limitations related to them. In this thesis, each three types of the real-time accident duration prediction model and real-time queue length prediction model is analyzed using artificial neural network optimized by genetic algorithm. The each prediction result is compared with actually calculated accident duration and queue length which is obtained from actual accident data and traffic flow data at the time of traffic accident occurrence. The traffic accident data is obtained from the Traffic Accident Surveillance and Analysis System (TASAS) and the traffic information data is obtained from the Freeway Performance Measurement System (PeMS). Two data are combined to predict real-time accident duration and queue length, and the combined data is divided into training data set and test data set. The developed model is evaluated by various measurements and the possibility of applying the developed method is validated through the evaluation. The developed model shows that all the real-time accident duration prediction models are predicted more accurately with lower Mean Absolute Percentage Error (MAPE) than all the real-time queue length predication models in both the training data set result and the test data set result. The results have brought the applicability...