Back-propagation neural network model to predict visibility at a road link-level
Venkata R. Duddu, Srinivas S. Pulugurtha, Ajinkya S. Mane, Christopher M. Godfrey
Abstract
Fog and rain influence the ability of the drivers to see other vehicles, pedestrians, bicyclists, and fixed objects, and, hence, increase the likelihood of getting involved in a crash on a road. Resource constraints limit installing weather monitoring sensors at regular intervals (for example, every 1, 2, 3, or 4 km) on a road. Therefore, the objective of this paper is to develop a feed-forward back-propagation neural network (BPNN) model to predict fog / low visibility condition, and, apply at a road link-level. Five years of meteorological data were collected from 238 weather monitoring stations in and around North Carolina to develop and validate the BPNN visibility model. The predicted visibility using meteorological data was compared with visibility from the weather monitoring sensor installed at five geographically distributed locations in North Carolina. The BPNN visibility model has better visibility prediction capability during a normal foggy condition when compared to visibility during thick/dense fog condition.