Proactive travel time predictions under interrupted flow condition

This research is focused on the development of a model for estimating arterial travel time by utilizing Automatic Vehicle Location (AVL) system-equipped bus as a probe vehicle. As an initial achievement, a prototype arterial travel time estimation model, applied to the bus arrival time estimation, was developed. The methodology adopted in this phase of the travel time estimation model was the on-line parameter adaptation algorithm. Three objectives were identified for this phase of the research. These were: 1) studying dynamics of bus behavior at a single bus stop, 2) extending the dynamics of bus behavior study to multiple bus stops, and 3) developing a prototype bus arrival time prediction model. The prototype travel time estimation was tested and evaluated through the simulation. INTRODUCTION Deterministic model of bus operation was first introduced by Newell and Potts [l]. Bell and Cowell [2] suggested the more descriptive dynamic model which covers bus journey times between the single bus stop and multiple bus stops, and expanded Newell and Potts’ model by introducing recursive autoregressive model. However, one of the unrealistic assumptions that both of those former researchers had made was that the passenger arrival rate at bus stop and boarding rate were time independent values. In reality this assumption is not valid and, therefore, it is assumed in this research that passenger arrival rate and boarding rate are time dependent. A prototype model development in this research consisted of tliree tasks. The first task was focused on the study of dynamics of bus behaviors at a single bus stop. Number of boarding passengers were simulated based on time varying passenger arrival rate and boarding lime. The second task was based on the study of travel time estimation at multiple bus stops. The dynamics of bus behaviors at multiple stops were simulated based on ratio between passenger arrival rate and passenger boarding rate. The main variable focused on this simulation was the departure time headway. Bus bunching is discussed at the end of t h s task. Finally, arrival time prediction model based on parameter adaptation algorithm [3,4,5,6,7] was developed. In this model, least square parameter adaptation algorithm [7] with forgetting factors was adopted. Two simulations scenarios, one with constant and the other with varying parameters at each bus stop, were tested in order to identify the parameter update capability. The prediction model was analyzed according to parameter errors and estimation errors. Currently, discrete time version of SI iding mode parameter estimation algorithm is being developed, for sliding mode technique can more effectively accommodate varying parameters. We will next study the impact of signalized intersections on travel time prediction, which will be a subject of f u t m paper. SING~LE Bus AT A SINGLE Bus STOP Model Formulati{on The initial approach of development of travel time estimation model in this research relies on the extended autoregressive model of Bell and Cowell 121. The characteristics of autoregressive model is that the most recent output affects the current status of model the most through the adaptive process. The enhancement made in this research was the adoption of time varying passenger 0-7803-2587-7195 $4.00