Due to the increased demand for reliable Advanced Traveller Information Systems (ATIS), accurate travel time prediction is gaining more and more importance. This has resulted in the development of many travel time prediction methods in recent years. However, in order to be able to provide travellers with individual travel time information, a number of improvements are still needed. First of all, there is often only limited information for a limited part of the route. Next, in cases of incidents, where the need for information is becoming more urgent, the provided information to travellers is often quite limited. Thirdly, the information to the travellers has to become more personal, more complete and more predictive. Reliable travel time information can compensate in a number of ways for these shortcomings. For that aim, an open framework for travel time prediction, PredicTime, is being developed. PredicTime will use the predictions of several, separate forecasting methods to predict future travel times from door to door. It should in the end be able to forecast travel times for virtually any route for each possible time of departure. It should take into account the actual traffic situation, the consequences of incidents, events, road works, etc. Finally, it should be open to new data sources like floating car data, and future prediction methods. Based on a review of the state of the art of travel time prediction methods, it can be concluded that indeed no single method is expected to be the best method under all circumstances. Instead, each method seems to have its own strenghts and weaknesses. If the full prediction horizon is to be covered, a combination of methods seems to be the best option. This idea has been incorporated in the functional architecture of PredicTime. For example, a Method Jury will be developed to 'judge' and combine the predictions made by individual travel time prediction methods using data fusion techniques in order to obtain a better forecast. Besides the Method Jury, the focus will also be on the development of a number of travel time prediction methods. These prediction methods should be able to predict travel times for a link or segment in the network, and they should also be suited for application at a network-wide level (taking into account tuning and recalibration requirements). (Author/publisher)
Abstract