A estimação neural de tempos de viagens de ônibus sob regime de fretamento usando-se de dados de posicionamento por satélites (GPS)

Abstract

The research's objective was to develop a model of bus arrival prediction to employees' transportation with charter regime. For this, was used a mathematical model based on Artificial Neural Networks. Previous experiments using RNA application to modeling this kind of problem considered as data's input "static" information of the route, such as the position of points of predetermined stops, the distance between them and the average speed route previously analyzed. The research's contribution was to develop a learning model considering as input data, any vehicle location (longitude and latitude) along the route and the time in which it went through the same coordinated, becoming the most dynamic and simplified model to be reproduced and incorporated into vehicle tracking tools that use global positioning (GPS). Another methodological contribution was structuring a "test tree", this test considered variations in the typology, in the values ​​for Learning Rate and "Momentum". The result achieved reached a predictive model with mean relative error of less than 1%.