Increasing congestion is becoming more and more a big issue nowadays. The results of this increase have negative effects on the environment, economy and safety. A way to decrease congestion can be found in using the maximum capacity of road networks by means of dynamic traffic management (DTM). Short-term congestion prediction should be a powerful tool in this application because - unlike the present DTM where actions are taken after traffic is already in a congested state - it can provide information about where and when congestion is to be expected, and the necessary operational management actions can be taken before this situation is reached. This paper aims to investigate whether recently introduced techniques, called the Artificial Neural Network (ANN) techniques, have possibilities in traffic and transportation science, and, more specifically, congestion predicting. Advantages of ANNs include short development phase, robustness, computer time efficiency, and adaptability to environmental changes. There is, however, a negative aspect (the so-called `black-box' aspect) consisting of the inability to analyse what is going on `inside' the ANNs, which can somewhat be minimized by a technique called `dithering'. The conclusion is that ANN techniques have possibilities in congestion prediction, but more thorough research is needed to compare the ANN techniques with more conventional methods, for instance, time series analyses. (A)
Samenvatting