Investigation of Atmospheric Effect and Road Damage on the event of Accidents Using Artificial Neural Network Model

Because of the infrastructure and critical role of roads, the displacements made in them have an enormous impact on the economic growth of countries. In recent years, road accidents have increased dramatically due to the lack of standardization of vehicles and existing roads, environmental factors as well as roadside contributors. Failure to observe safety principles in geometric design and maintenance plan of roads has caused serious damages to the country. On the other hand, modeling of crashes and examining the variables affecting its occurrence has not been done to provide corrective options. Therefore, in this paper, with the help of the intelligent artificial neural network model, the simultaneous effect of atmospheric factors and road pavement collapse was investigated and modeled on the occurrence of accidents on the busy routes of Khuzestan province. Meteorological data and accidents statistics for the statistical period of 2011-2014 were prepared. Independent and dependent variables, as inputs and outputs of the neural network, were defined at three levels of total accidents, damage accidents and injuries accidents. The modeling was conducted using the Neuro Solution software as leading and based on the Levenburg-Marquardt training principle and Sigmoid axon transfer function. The results of various architectures showed that the designed neural network, with high explanatory factor, can model the total number of accidents. In this model, the most effective parameters on the creation of accidents were weather conditions (rain and cloudy weather), road failures (PCI index and prevailing failures), and another factor (the number of forward-facing crashes) as model inputs. For modeling of damage accidents, effective inputs included the number of accidents (damages and losses), road failures statistics (PCI paths and prevailing failures) and weather conditions at the moment of accident. The achievement of the high-level explanatory factor in this model also reflects the ability of the architectural neural network to model the crashes. The obtained results show that using the variables introduced, prediction of accidents can be used in terms of the rate and severity of accidents (incidence, fate or damage). Also, the impact of atmospheric factors on the occurrence of accidents is an undeniable fact, and it is necessary to modify the routes, increase road safety and inform the drivers, to reduce accidents during the occurrence of precipitation or frost and mist conditions