MODELO SISTÊMICO PARA DETECÇÃO DE PONTOS CRÍTICOS EM RODOVIAS COM BASE EM SEGMENTOS HOMOGÊNEOS

Abstract

The study presents an innovative approach to defining homogeneous segments and identifying critical locations on highways, improving the accuracy of hazardous point analysis. Unlike the traditional DNER (1986) methodology, which uses 1 km segments, this new approach proposes a more detailed segmentation with 50-meter sections, allowing for a more refined analysis of the physical, operational, and geographical conditions of roadways. The research was conducted on the MG-050 highway, using accident data from 2016 to 2022, along with information on road type, gradient, signage, and traffic volume. Geographic Information System (GIS) tools were employed to create thematic maps that aided in visualizing critical points. Indicators such as Standard Gravity Unit (SGU), Hazard Index (HI), and Critical Index (CI) were used to assess segment criticality. One of the main methodological innovations in the homogeneous segmentation process was the application of the XGBoost algorithm, which demonstrated high efficiency in distinguishing between critical and non-critical segments. Trained with variables such as traffic volume, weather conditions, and accident history, the model achieved average indicators above 0.85. The results identified a total of 1,203 segments using the new methodology. The thesis also contributed to the SGU indicator calculation process by segregating accidents involving injured victims inside and outside vehicles, as well as defining parameters for weighting vehicles with victims, considering the occupancy rate of a standard vehicle.