The analysis of mobility data is complex as they can be represented as a high-dimensional space where data are linked temporally and spatially. It is therefore necessary to manage both a high number of raw sequence data (time series) and to keep all their relevance from a geospatial point of view.
Challenges are being accentuated by the existence of many data silos (systems not interconnected), the rapid growth of sensors data (IoT) and the influence of external factors (meteorological conditions, events, road works ...)
The smart mobility revolution is also tightly associated with the rapid rise of electrical vehicles (EVs), cheaper and higher-capacity batteries and the rapid progress of distributed power generation. Machine learning models for smart mobility need therefore to take into account energy data (demand / production / storage).
Urban Transit AI use cases
Real-time delay management
Urban transit is heavily dependent on external events. So far most urban transit systems fail to communicate on the real impacts of these events on the commuter journey (delays particularly). A new generation of ML-based models can be developed to implement smarter predictive models and personal notifications.
Urban transport demand forecast
Demand forecast is a core requirement to optimize routes, schedules and capacity. The digitalization of transport operations (e-ticketing, mobile, IoT…) generate large sets of data that can be consolidated and enriched to feed ML-based forecast algorithms.
ML models for parking places demand forecast can be used to implement services related to real-time routing of drivers and riders, modular pricing systems implementation and multimodal transportation services.
Charging Stations / Grid Management
Demand forecast for EVs charging stations
Electrical Vehicles owners and users of EVs sharing services (bikes, scooters, cars) are on the rise and generate new consumption patterns related to both energy demand and parking space demand. Charging stations management requires new ML models to optimize pricing models and demand forecast.
Solar power forecast for EVs charging stations
Solar power generation is becoming a key component of distributed generation. ML models can help anticipate how solar powered charging stations can accommodate EVs energy needs and how they can be integrated into the local distribution grid.
Smart Grid Management
EVs power demand can put some pressure on the local distribution grid. At the same time some energy stored in EVs could be fed back into the power grid (Vehicle-to-Grid / V2G) when vehicles are not in used. ML models can help manage the integration of EVs into smart distributed energy resources systems. The use of EVs as temporary stationary storage units is still at an early stage of development but it is a promising one. It is definitely a mean to sustain the growth of the circular economy through the optimization of renewable energy storage.
Autonomy Management (battery charge)
Telemetry systems give the capacity to acquire real-time data on battery levels, driving behavior and vehicle parameters. ML models can in near real-time predict the state of charge (SOC) of EVs batteries according to the anticipated route of the driver.
Dispatching / Routing optimization
Routing and dispatching are complex operations. For an EVs fleet the models are even more complex as the anticipated autonomy of vehicles and the time to charge batteries need to be integrated too. This is a case where deep learning has shown interesting results.
ML models can be trained to anticipate when a battery should be changed in order to keep good level of performances for EVs. This type of model can be trained for other components of the vehicles through the acquisition of sensors or bus CAN data.