MOBILE PHONE Data in transport

In the next step, the naïve OD demand was updated such that the difference between modelled and actual flows is minimised using an artificial intelligence based optimisation algorithm: the Genetic Algorithm (GA). A parallel implementation of GA was run on CDAC’s computing cluster with an open-source traffic modelling software called SUMO used for traffic assignment. The overall estimation method is shown in the figure below.

Tapas Saini
(C-DAC)

Traffic counts on a number of points on roads going in and out of South Mumbai (cordon counts) that were not used in the above calibration process was used to independently verify the accuracy of the estimate. Comparison between modelled and actual flows at cordon points yielded a Mean Average Percentage Error (MAPE) of 9.69%. This is in line with accuracy expected using traditional transport demand modelling methods.

Advantages of Mobile Phone Data

The method used in the proofof- concept study has a number of advantages over the current state-ofpractice in demand modelling. The mobile phone data based approach can be used to estimate temporally rich mobility patterns, or dynamic OD matrices, that captures variations in mobility patterns during the course of a day. Traditional demand modelling methods approximate the extra demand during peak times using a simple peak factor. The OD matrices can be re-estimated periodically, for example every month, to understand seasonal variations in mobility patterns within each year; this is currently not possible using traditional tools and methods. Moreover, it was also demonstrated that the method used in the proof-of-concept study can also be run on a cluster of desktop PCs using open-source tools.

Rajesh Krishnan
(ITSPE)

Overall, up-to-date dynamic OD information is invaluable for transport planning and operations in our fast growing urban areas. The proof-ofconcept project has demonstrated that such information is within reach of the stakeholders without having to make large investments in computing power, software or manual surveys.

This article is based on the work of a joint CDAC-ITSPE team. The authors would like to acknowledge Shikha Sinha and Venkata Srikanth from CDAC and Rajesh Gogineni and Rakesh Behera from ITSPE for their contribution to the project. Above all, the author would like to thank Mr V. Muralidharan from CDAC for acting as a catalyst and inspiration behind this project.