Self-adapting traffic flow status forecasts using clustering

A key problem for the efficient use of stationary traffic prediction models is that for adaptation to new data they require human-made re-calibration with a new database. So far, there has been a lack of knowledge of how to develop a practical prediction model that would learn while working online. Anyone providing real-time traffic information and making forecasts of the traffic situation could benefit from such models. The aim of the study is to develop a method to make a self-adapting short-term prediction model for the status of traffic flow. The principles for such a model are described. The method is based on self-organising map and the model is implemented on a highway in the Helsinki Metropolitan Area. Specifically, the structure of the model makes it possible for the model to learn by itself without the need to save all the data into databases. Consequently, long-term online use of the model makes fewer demands on computers. The results indicated that the self-adapting principle improved the performance of the model. The principles of the model can also be applied in other locations.