Development of Speed Distribution Model and Prediction of Geometric Parameter Influence on Driver Speed through Artificial Neural Network

抄録

The safety of a highway can be assessed easily using speed consistency criteria. The 85^th percentile speed (V₈₅) values used to measure consistency is obtained from speed prediction models. So far, the majority of speed prediction models are regression models. Machine-learning techniques have recently been explored to develop speed models but have not yet fully explored. Further, it is also known that speed distribution plays a vital role in V₈₅. Thereby in this study, a speed distribution function is developed using Back Propagation Neural Network with naturalistic speed data. The developed model yields better results with R² of 0.91 and RMSE of 2.8 in predicting various percentile speeds on speed distribution at the center of the curve. The profile method sensitivity analysis conducted on the developed model revealed that the parameters superelevation and radius have the most significant influence on the speed over the curve. One of the significant advantages of the developed model over the existing speed models is its capability of predicting any percentile speed.