Most utility integrated resource plans model EV load growth as a gradual, smooth addition to the existing load profile — a reasonable simplification for long-term capacity planning. It is the wrong model for short-interval load forecasting, where the operationally relevant characteristic of EV charging is its spatial and temporal concentration, not its aggregate MW growth.
The Distribution Problem
Statewide EV adoption projections from the DOE, EPRI, and regional grid operators typically report aggregate charging load in MW at the balancing authority level. These projections are useful for multi-year capacity planning. They obscure the operational reality that EV charging load at the distribution feeder level is highly non-uniform in both geography and time.
EV ownership concentrates in specific demographics and geographic areas: higher-income households, urban and suburban areas with access to Level 2 home charging, and transit corridors with DC fast charging infrastructure. Within a 1,000 MW balancing authority territory, the practical EV charging load may be concentrated on 15–20 distribution feeders serving these areas — creating localized load spikes that are invisible in aggregate telemetry but substantial at the feeder level.
For a utility with feeder-level metering, this creates two distinct forecasting challenges: (1) aggregate system load forecasting must account for the new EV charging component, and (2) feeder-level forecasting must model the specific charging patterns on affected feeders, which differ structurally from legacy residential load patterns.
The Evening Charging Spike and Its Timing Uncertainty
L2 home charging (7.2–19.2 kW per vehicle) produces a characteristic load pattern: vehicles arrive at home typically between 5:30 and 7:00 PM, connect to chargers, and begin drawing load immediately unless managed by time-of-use pricing or utility-controlled charging programs. The resulting load spike coincides with the traditional evening peak and adds to it — a superimposition problem, not a new separate peak.
The timing uncertainty is significant for short-interval forecasting. The arrival time distribution for EV charging load is wider than the arrival time distribution for traditional residential HVAC cycling: charging start times vary by 90+ minutes depending on commute patterns, which cannot be predicted from the same weather and time-of-day features that predict HVAC load reasonably well. This means EV charging load contributes disproportionately to forecast error at the 15-minute horizon compared to its contribution to aggregate daily energy demand.
Managed charging programs — utility or aggregator-controlled systems that delay charging start times to off-peak periods — address the timing problem but create a new one: the managed load becomes a controllable variable whose behavior depends on the program's dispatch decisions, not on observed physical patterns. Forecasting managed EV charging load requires knowledge of whether the program will dispatch, at what time, and with what participation rate — essentially a joint problem of load forecasting and demand-response program performance forecasting.
DC Fast Charging: A Different Operational Profile
DC fast chargers (50–350 kW per port) at commercial locations have a fundamentally different operational profile from home L2 charging. A 10-port DCFC station can draw 1.5–3.5 MW when fully utilized. Station locations tend to cluster along highway corridors and near destination locations (shopping centers, restaurants). Utilization peaks during highway travel hours — midday and late afternoon on weekdays, midday through early evening on weekends — not during the traditional residential evening peak.
For distribution system operators, a new DCFC station represents a step change in feeder load that may approach or exceed the feeder's historical peak. Standard forecasting models trained on historical data will not predict this step change — it's a discrete infrastructure event, not a gradual trend. Utilities need to track planned DCFC installations in their territory and incorporate them as known load additions in the forecasting system, similar to how large new industrial customers are handled.
The SCADA data quality issues discussed in our article on stuck registers and SCADA data quality failures are particularly relevant for DCFC installations: a DCFC station that charges from near-zero to 2+ MW in under a minute will produce rate-of-change values that look like step changes in the validation logic. Forecasting systems need DCFC location metadata to distinguish legitimate DCFC load ramps from data quality anomalies on the same feeder.
Vehicle-to-Grid and Its Forecasting Implications
V2G (vehicle-to-grid) capable vehicles and bidirectional chargers create the theoretical possibility of using parked EV batteries as grid storage — discharging during peak demand and recharging during off-peak hours. Several utilities are piloting V2G programs, particularly in California and Hawaii.
For load forecasting, V2G introduces a new category of uncertainty: enrolled vehicles can be net consumers or net producers of electricity at the same physical location, depending on program dispatch decisions. A feeder with 50 V2G-enrolled vehicles may show net positive load during off-peak hours (charging) and net negative load during peak hours (discharging) — a pattern that requires explicit modeling of the V2G program's dispatch rules, not just historical load regression.
At current enrollment levels, V2G's operational impact is below the noise floor for most utility service territories. However, utilities in early-adopter markets should begin tracking V2G enrollment and program dispatch records now, before enrollment scales to the point where unmodeled V2G behavior produces systematic forecast bias.
What Utilities Should Do Now
The practical steps for utilities anticipating significant EV load growth in the next 3–5 years are: (1) identify which distribution feeders currently serve the highest concentrations of EV-owning customers (EV registration data, utility billing data, and census income data can all serve as proxies); (2) install interval metering on those feeders if it isn't present already; (3) begin tracking DCFC permit applications and interconnection requests as planned load additions in the forecasting system; and (4) evaluate whether time-of-use rate structures or managed charging programs are feasible, given that managed charging converts an unpredictable load into a partially controllable one.
The aggregate EV load growth number — total MW added to the system by year 2030 — is the input that resource planners need. What operations and forecasting teams need is the distribution of that load across feeders, the timing distribution of charging events, and the rate at which the load profile is changing as enrollment grows. These are different questions that require different data collection programs.