Why EVs Make Availability Harder
A vehicle can be at the depot yet still be unavailable when demand rises. In an ICE fleet, availability mainly means the vehicle is mechanically usable, and refueling a full tank takes three minutes. For EVs, it also depends on SoC, charging status, expected charge completion, and whether the occupied depot slot will be free for the next high-demand trip window.
An EV parked at the depot but needing two more hours to reach 80 percent SoC is unavailable for the morning peak. Knowing this at 04:30 helps. Discovering it at 06:45, when the driver needs the vehicle, does not.
What "Available" Means in an EV Fleet
Before creating a forecast, define availability around EV operating limits. The definition used here has three parts:
- SoC threshold met. Battery at or above the minimum state of charge required to complete the expected next trip plus a return buffer. For a 25-km Jakarta ride-hailing shift, a working threshold is 55 percent or higher, which provides margin for traffic delays and unplanned detours.
- Ready time alignment. The vehicle finishes charging, or is confirmed to already have sufficient charge, by at least 30 minutes before the next expected dispatch window.
- No scheduled maintenance block. The vehicle is not flagged for service or inspection during the forecast period.
Meeting all three conditions means the vehicle is operationally available. Meeting only two signals planning risk. At key checkpoints, the share of fully available versus partially available vehicles is more useful than a basic "vehicles at depot" total.
Data Needed for Availability Forecasting
Fleet-scale availability forecasts do not need machine learning. A structured calculation using dependable inputs is enough. The important inputs are:
- Current SoC per vehicle, updated at trip end or at a regular interval from the telematics feed
- Current charger connection status: plugged in, charging actively, or idle at port
- Charger power output in kW for the port the vehicle is connected to
- Vehicle battery capacity (usable kWh) from the specification table in your fleet system
- Target SoC and target ready time from the overnight charge schedule
Expected charge completion is easy to calculate from these inputs: target SoC minus current SoC, multiplied by battery capacity, divided by charger power output, then adjusted for normal charging efficiency. At 30 percent SoC, a 30-kWh battery on a 7.2-kW AC charger reaches 80 percent in about 2.1 hours under normal conditions. The math is simple. Reliable data is the harder part.
The Key Forecast: 07:00 Readiness
A 12-vehicle Jakarta ride-hailing fleet with 5 vehicles still on AC power at 07:00 is effectively a 7-vehicle fleet for the first hour, while carrying the fixed costs of 12. For these fleets, the most important availability forecast is which vehicles will be ready by 07:00. Morning demand is concentrated, predictable, and unforgiving.
The 07:00 readiness forecast should be available by 23:00 the night before, using overnight charging schedules. A vehicle scheduled to finish at 07:20 creates a gap that can be addressed eight hours early: move it to a faster charger, change its charging start time, or assign it to later morning trips. At 23:00, corrective choices remain. At 06:45, they do not.
In early pilot work with Jakarta-area fleet operators, running the forecast at night and reviewing it before coordinators left the depot materially improved 07:00 availability versus the former practice of checking charger status at 06:30 and reacting afterward.
Where Forecasts Fail
A charging forecast is accurate only when the vehicle charges as scheduled. Charger faults, partially inserted cables, and vehicles returning later than expected can all quietly invalidate the model's assumptions.
The failure mode is a forecast that says 11 of 12 vehicles will be ready by 07:00 because it assumes VH-08 is charging at full rate. In fact, VH-08 has been at a faulted port since 01:00 without receiving energy, while the charger's status updates stopped at the fault. Without an alert for projected SoC that stops rising during a scheduled charging window, the problem stays hidden until it matters.
A forecast without anomaly detection is unfinished. The availability model also needs to identify vehicles whose charging progress has fallen behind schedule. An alert at 03:00 still leaves time to move the vehicle to a working port, turning the forecast into operational value instead of false confidence.
Scale the Approach Without Overbuilding
When building an availability forecast, it is tempting to include everything: demand patterns, traffic delays, driver availability, maintenance schedules, and battery degradation. For fleets under 20 vehicles, that is overengineering. With limited data, multi-variable models may look numerically precise but often prove wrong because uncertainty compounds across variables before they can be reliably calibrated.
A more practical approach forecasts what dependable inputs can support, especially SoC-based charge completion, and manages the rest with buffer. Add slack to the readiness threshold instead of modeling uncertainty with false precision. A 55 percent minimum SoC threshold includes implicit protection for normal traffic variation. A traffic model producing a threshold to the nearest kilometer suggests precision that the data cannot support at this fleet size.
At RideBlitz, the availability model built for early-access fleet operators begins with SoC-based charge completion and adds two practical layers: charger port monitoring to detect silent faults, and a late-return flag for vehicles missing their expected depot arrival time. Both are simple to implement and catch most availability surprises before they affect the morning peak.
