The purpose of this research was to study the influence of various factors on the results of the short-time prediction of the traffic situation on motorways. The models were made to predict the speed and flow 15 minutes ahead of the observation period in five-minute periods. Multi-layer perception (MLP) networks were used as prediction models. According to this study, it was better to increase the number of hidden neurons by reducing the input parameters by decreasing the number of cross-sections rather than by shortening the time-series. The models that were divided into two sub-models - one for the mean speed forecasts and the other for the traffic flow forecasts - gave better results than one single model predicting both variables simultaneously. For 90 percent of the predicted flows the relative error was 20 percent at most, and for 90 percent of the predicted speeds it was four percent at most.
Abstract