BPNN (ANN) Based Operating Speed Models for Horizontal Curves Using Naturalistic Driving Data

Abstract

Safety is a major concern when dealing with the geometry of the road. To improve safety, designers started using operating speed prediction models in a geometric design. So far, the majority of the available works on speed models was focused on 85th percentile speeds, giving less importance for the rest of the percentile speeds models. However, the other percentile speeds such as 15th, 50th, 95th, and 98th do exhibit their influence on geometric parameters in geometric design. These percentile speeds needed further exploration. Thereby in this study, percentile speeds models are developed using Back Propagation Neural Network (BPNN) with naturalistic speed data collected on horizontal curves. The percentile speed (Vp ) model developed yields better results with R2 of 0.83. The developed model also showed that design speed has the most substantial influence on percentile speeds with the Relative Parameter Influence (RI) of 16%. The percentile speed results obtained from the BPNN model show normal distribution (K-S test). We can say that the developed model represents the naturalistic free-flow speed distribution. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.