Average speed prediction using artificial neural networks

title: Average speed prediction using artificial neural networks
author: Cynthia Jacobs
published in: August 2003
appeared as: Master of Science thesis
Knowledge Based Systems group
Delft University of Technology
PDF (3.252 KB)


Nowadays cars have navigation systems guiding the traveler from departure to destination. The navigation system uses route planners to determine the best route for the traveler. However these route planners do not use dynamic traffic information to decide what route is best. These route planners use the maximum speed allowed on a road to determine the travel time. So it is possible that the traveler is guided over congested routes while another route will lead the traveler faster to its destination.
In this thesis at first it is explored what influences the average speed on the road and it is tried to predict this average speed using a feed forward neural network. If this average speed is known the real travel time can be determined for parts of the road and the best route can be discovered using this travel time.
Analyzing the data showed that time, day of the week, month, weather, events, holidays and special events like accidents are factors that influence the average speed on the road. These influencing factors are used to predict the average speed on a specific point on a road. Therefore three models are developed. Two of these models are tested using JavaNNS developed at the Wilhelm-Schickard-Institute for Computer Science in Tübingen, Germany.
The research on the architecture of neural networks and tests showed that a feed forward neural network with one hidden layer, from which the number of hidden neurons is determined by the formula of Fletcher and Goss (between 2Ni+1 and 2vNi +No), is capable of predicting the average speed using only the time, month, day of the week and the average speed of 40 minutes before. At least 76% of the predictions have a difference smaller or equal to 10% if the prediction is made 40 minutes ahead.
From these results it can be concluded that Artificial Neural Networks are capable of making good approximations for problems like predicting the average speed.

blue line
University logo