the project relies on three keypoints:
The prediction system relies on historical data bike usage provided by Intermobility, weather conditions and local holiday dates to generate bike usage prediction for each considered bike network. The models for bike usage prediction are generated by supervised machine learning algorithms. In the context of this project, we notably use Random Forest and Long Short-Term Memory algorithms to obtain the best possible prediction accuracy for different horizons.
The balancing system relies on the current state of the bike network and on the predictions generated by our algorithms. Upon request, the system is able to provide optimal routes and operations for the teams that perform the rebalancing operations; these operations will ensure a sufficient number of bikes in each station of the network to fulfill the future demand.
The visualization plateform allows bike network managers to monitor the current status of his bike network(s) as well as the predictions of bike usage at each station for different time horizons (up to 12 hours ahead). Additionally network managers are able to manage rebalancing missions performed by his teams on the field. Upon request, the produces optimal missions (routes and operations at each station) to equilibrate the number of bikes at each station according to pre-defined target thresholds and boundaries. The teams on the field automatically obtain clear instructions on their tablet on the tasks that they must perform.