Procedimento para uso de Redes Neurais Artificiais no planejamento estratégico de fluxo de carga no Brasil

This paper aims to use the technique of artificial neural networks (ANN) to estimate the origin-destination matrix (O-D) of soybeans in Brazil for export to explain the variability of flows between O-D pairs, considering the dynamic characteristics existing in O-D matrices. We will then compare the results with the gravity model (GM), which is a model used in strategic planning of the brazilian government also proposed a procedure for the use of anns in travel distribution load. among the four models built in this paper, we have highlighted the combination C-02, with a coefficient of determination (R²) higher than 9% and with a dissimilarity index (DI) 6.92% lower than the GM. It can be observed that anns can be a potential substitute to conventional statistical models because if their easy interface to user programs, but the ANNs are not reported in the literature as options for calculating the travel distribution of Four Steps Method. Thus, this technique is rarely used for the transportation of cargo.

neural networks; soybean exportation flows; cargo flow prediction; trip distribution