Context
Electric vehicle (EV) charging demand is becoming an increasingly important input for electricity procurement and grid planning for utilities and charge point operators (CPOs).
Challenge
Utilities and CPOs lack visibility on EV charging demand, leading to inefficient electricity procurement, grid stress, and inability to anticipate peak loads.
AI Solution
The solution uses ML for data aggregation from multi‑source EV charging consumption data, spatial‑temporal modelling, localized forecasting at the charging‑station level, and decision‑support integration to enable operational planning beyond standalone prediction. The model forecasts EV charging consumption using historical usage, location data, number of chargers, and temporal patterns and produces short- to mid-term forecasts (up to 3 months).
Impact & evidence
Impact is evidenced by forecast accuracy levels of around 15% Mean absolute percentage error (MAPE), improved performance compared to baseline forecasting approaches, the ability to anticipate demand peaks, and active operational use of forecasts by planners.
Implementation / Adoption
Implementation requires the availability of historical EV charging data, asset metadata for charging stations and chargers, and external signals such as time and location. Deployment followed a staged approach, starting with an MVP developed over 6–10 weeks, a pilot deployment lasting 2–3 months, and subsequent industrialisation over a 6–12 month period to enable broader operational use.
Data
The solution requires charging consumption data, station locations, number of chargers, time variables (hour/day/seasonality), cost of recharging, holidays and weather.