Charge · Charging infrastructure & traffic

ChargeCast

Traffic sensors already know where drivers are heading, often hours before they plug in. ChargeCast turns live traffic flow into a 1–3 hour forecast of charger occupancy for every hub in the city, so drivers get sent to a free plug. It also logs every driver who found a hub full and clusters that unmet demand to show where the next charger should go.

Synthetic city · 10 weeks of traffic + 5-minute queue simulation · last 2 weeks unseen by the model

Now

forecast occupancy at +h unmet demand (last 24 h) proposed site dots = traffic sensors all unmet demand (2 weeks)

Find me a free plug

Click anywhere on the map to set your position.

I'll arrive in

No position yet.

Hub

Where to build next

weighted k-means on drivers turned away

What the forecaster looks at

Why it matters

Drivers' biggest complaint is arriving at a charger that's occupied, and operators' biggest risk is building one in the wrong place. Both are prediction problems. Traffic data already exists in every city, so this approach needs no new sensors.

Method

Traffic → arrivals: each hub's arrival rate depends on kernel-weighted nearby flow, its venue type and the time of day.
A random forest (50 trees, 20 features) predicts occupancy 1, 2 and 3 h ahead. It's compared with "same as now" and "same as last week".
Siting: drivers who found a hub full (plus demand in areas with no charger) are clustered with weighted k-means. Each centroid is either a new site or a "+plugs" upgrade.

Limits & next steps

Synthetic city. Next: real open traffic-counter feeds and charge-point status data (OCPI).
"Chance of a free plug" assumes independent plugs; a queueing-theory estimate would be tighter.
Next: route-aware demand (origin–destination), and grid-capacity constraints for siting.