Interesting post, and it's very nice that the author decided to share all the scripts. Forecasting the demand is the first fundamental step to actually optimize the rebalancing operations that usually happen overnight. There is a lot of academic and applied research in this field.<p>Prof. David Shmoys (Cornell University) is working on optimising the rebalancing operations of the of CitiBike NYC (press article <a href="http://www.news.cornell.edu/stories/2015/01/cornell-research-steers-nyc-bikes-needy-stations" rel="nofollow">http://www.news.cornell.edu/stories/2015/01/cornell-research...</a>), Prof. Günther Raidl has been doing the same for a while for the CityBike Vienna (Austria) network (<a href="http://www.citylab.com/commute/2014/08/balancing-bike-share-stations-has-become-a-serious-scientific-endeavor/379188" rel="nofollow">http://www.citylab.com/commute/2014/08/balancing-bike-share-...</a>), and three researchers from Udine (Italy), and Vienna (Austria), including myself, have been working on the same problem on similar data sets (summary paper <a href="http://link.springer.com/article/10.1007/s10601-015-9182-1" rel="nofollow">http://link.springer.com/article/10.1007/s10601-015-9182-1</a>, or preprint <a href="http://www.tunnuz.net/documents/digaspero_rendl_urli_constraints2014.pdf" rel="nofollow">http://www.tunnuz.net/documents/digaspero_rendl_urli_constra...</a>).