Transport Models and Big Data Fusion: Lessons from Experience
Luis Willumsen, Kineo MObillity Analytics, Miguel Picornell, Kineo MObility Analytics
The use of new sensor data (GPS, mobile phone, Bluetooth) has been proposed as a solution to travel data collection. The paper explores technical issues involved in using these data sources.
Data from a large number of new sensors has been proposed as providing new and richer data sources to improve transport models and forecasting. Bluetooth, GPS, Twitter, WiFi and mobile phone data has the potential to strengthen the data for our models together with the promise of improved and more reliable forecasting.
The paper will provide a summary of the strengths and limitations of each of these data sources and focus more closely on the use of mobile phone data. Anonymised mobile phone data offer a great potential as being able to identify a trip from true origin to true destination without requiring much processing by the operator. The use of this type of data has resulted in mixed results so far. We, at Kineo, have been processing this type of data in a couple of countries and assisting in the calibration of transport models based on this source. We have learnt a good deal in the process.
The most important lesson is that processed anonymised mobile phone data is not directly transferable to a transport model; and there are some very good reasons for this.
For a start, the original mobile phone data is often of a greater granularity than the traffic and travel data collected by conventional methods: road side interviews, household surveys. The complete network is available for these trips whereas most transport models use a sub-set of main roads in most cases. Aggregation in these cases is no trivial. Other constrains depend more closely on how exactly the mobile phone company processes the operational data used; these, again, influences issues like expansion and data fusion. We will provide some examples of these issues and how they can be tackled.
The paper will summarise what we have learnt from this experience and suggest some guidelines for the successful use of this, and other type of sensor, data to improve transport modelling and forecasting.
Association for European Transport