01
Which customers are likely to adopt?
Estimate household-level ownership probability from income, education, commuting, housing and other characteristics.
AI-Assisted EV Ownership and Hourly kW Forecasting
Convert future EV adoption into the hourly loads utilities must plan for.
GIM estimates which representative households are most likely to own EVs, develops charging requirements from travel and charging assumptions, and adds those loads to existing whole-home 8,760 profiles.
Customer ownership probabilities · Charging diversity · Whole-building and localized hourly loads
The missing connection
EV counts do not tell a utility the kW impact
The same number of EVs can produce different grid consequences depending on where owners live, how far they travel, when they return, charger power, charging frequency and coincidence with existing household demand.
02
How much energy must be replaced?
Translate travel distance and vehicle assumptions into required charging energy rather than assuming every EV charges identically.
03
When does charging begin?
Use arrival timing and selected behavior assumptions to place charging within the customer’s hourly load profile.
04
How does charging combine with home load?
Add EV kW to air conditioning, heating, water heating, appliances and other coincident residential demand.
Customer-to-grid workflow
Build the kW forecast from the household upward
Each step preserves customer and time differences before results are aggregated to utility planning geographies.
Set EV growth
Select the service-area ownership scenario for the current year, 2030 or 2035.
Estimate ownership
Calculate an EV probability for each representative customer using nearest-neighbor analysis.
Develop charging
Convert travel, arrival and charger assumptions into customer hourly charging loads.
Add baseline loads
Combine EV charging with each customer’s existing whole-building 8,760 profile.
Aggregate and analyze
Produce customer, neighborhood, block-group, ZIP and service-area results.
Two linked model components
Separate the ownership question from the charging question
A household can have a high probability of EV ownership but create different hourly impacts depending on driving, arrival time, charging power and existing load. GIM models those relationships in sequence instead of treating adoption and peak kW as one average assumption.
The utility-wide EV scenario controls the overall ownership level. Customer probabilities determine where ownership is allocated, and charging profiles determine when the new demand appears.
Why GIM uses nearest-neighbor modeling
Flexible local relationships estimated with real data
The non-parametric nature of a K-nearest-neighbor (KNN) model can identify local ownership and commuting patterns missed by the more rigid parametric structure of traditional statistical models.
Practical advantages
The method works directly from observed similarities among households.
- Accommodates nonlinear and interacting customer characteristics
- Preserves local variation among household types
- Produces an intuitive customer-level probability
- Supports filtering and geographic aggregation after forecasting
- Can be updated as calibration data improve
How it works
Model estimation requires a reliable training" database including households with and without EVs.
- 26,000 households are in the GIM “training” database
- MAISY households are matched with training database households based on an algorithm including income, education, and other variables
- Ownership probability is calculated
- The matching algorithm is optimized
Intuitively appealing methodology:
KNN forecasts reflect the fact that "identical" households are likely to
have similar EV purchase inclinations. The modeling process reflects the fact
that the matching process is not perfect - that is, some household factors can't
be captured - so the forecast has a random element. That is why we estimate
the probability that each household owns an EV.
What shapes hourly EV load
Charger rating is only one part of the forecast
A Level 2 charger may draw 11.5 or more kilowatts, but the relevant planning quantity is the diversified, coincident charging load added to existing customer demand.
MI
Travel distance
Longer daily travel generally increases required charging energy and potential charging duration.
TIME
Return-home timing
Arrival patterns determine whether charging begins before, during or after the residential and G&T peak.
kW
Charger power
Higher charger capacity increases instantaneous load but may shorten the charging period.
HOME
Existing household load
EV charging must be evaluated alongside AC, heat, water heating, appliances and other customer demand.
DIV
Charging diversity
Not every EV starts charging at the same time, requires the same energy or remains at full charger output for the same duration.
MC
Program response
Time-of-use and managed charging can delay, stagger or shift demand, but can also create rebound or synchronized start effects.
Illustrative historical application
From individual charging diversity to neighborhood hot spots
The original Rhode Island example demonstrates why averages must be interpreted carefully. Individual customers show different charging start times and durations, lowering the average ZIP-level kW contribution relative to one charger’s maximum draw.
However, filtering the same area for higher-propensity customer characteristics revealed a neighborhood segment with substantially greater EV ownership and whole-home load impacts than the ZIP average.
- Individual charger demand does not equal diversified average demand
- A low ZIP-wide average can coexist with much larger neighborhood impacts
- Existing home load and weather change the total effect
- Customer filtering helps locate the segments most likely to matter
Historical illustration: The graphic reflects an earlier Rhode Island application and its assumptions. It should not be interpreted as a current forecast for Rhode Island or another utility.
Open the original EV charging kW graphic full size ↗Interpret the kW correctly
Four different load measures answer four different questions
Confusing charger nameplate kW with diversified or coincident EV load can materially distort both grid-risk and business-case results.
Charger nameplate kW
The maximum charging rate available from the equipment under applicable vehicle and circuit limits.
Charging EV kW
The load of an EV that is actively charging at a particular hour; it may be below nameplate.
Average kW per EV owner
Charging load averaged across all EV owners, including vehicles not charging during the hour.
Coincident system or local kW
The EV contribution during the utility, feeder, neighborhood or transformer peak being evaluated.
Whole-building kW
EV charging plus the customer’s baseline residential demand at the same hour.
Managed residual kW
The charging contribution remaining after participation, availability and control effectiveness are applied.
Current Grid Impact Model applications
Use customer hourly kW forecasts across technical and financial decisions
The ownership and charging models supply a common analytical foundation for the expanded GIM worksheets.
01
Localized EV forecasts
Compare customer, block-group, ZIP and service-area ownership and load impacts.
02
Transformer screening
Apply localized EV peak loads with utility-selected size, current loading and planning-limit assumptions.
03
Distribution engineering exports
Supply priority-area 8,760 profiles to CYME, Synergi, WindMil, OpenDSS and other workflows.
04
Managed charging scenarios
Compare unmanaged, time-of-use and controlled charging with participation and engagement assumptions.
05
Managed charging business case
Translate coincident kW reduction into avoided G&T demand charges, payback, benefit-cost ratio and NPV.
06
DSM, DER and VPP analysis
Evaluate EV charging alongside weather-sensitive loads, batteries and other flexible resources.
Responsible planning use
Detailed hourly forecasts improve screening evaluations
What the model adds
- Customer-level adoption probabilities
- Charging energy and timing based on travel-related inputs
- Whole-building 8,760 loads including EV charging
- Geographic and customer-segment aggregation
- Managed charging and mitigation scenarios
- Consistent inputs for grid and financial analysis
Additional application inputs
- Block-group results do not identify a specific asset without utility mapping
- Simulated transformer load impacts require inputs on transformer size, households/transformer and current tranformer loading
- Detailed engineering requires asset data, measurements and circuit studies
- Business-case inputs require current tariff and vendor information
Frequently asked questions
EV ownership and charging kW forecasts
Why not multiply EV count by charger kW?
Because not every EV charges simultaneously or at maximum power. Required energy, arrival time, charger limits and charging duration create diversity that must be modeled.
What does the KNN model forecast?
It estimates EV ownership probability for each identity-protected representative household by comparing that household with similar records in the EV ownership calibration database.
How is charging energy estimated?
GIM applies commuting distance and selected vehicle and charging assumptions to estimate the energy that must be restored and the hours over which charging occurs.
Why add charging to whole-building loads?
Transformers and feeders experience total customer demand, not EV charging alone. Air conditioning, heating and other household loads determine the actual coincident impact.
Can utilities change the charging assumptions?
Yes. GIM scenarios allow key EV growth, charging and program assumptions to be revised so utilities can compare alternatives rather than rely on one forecast.
Does the forecast replace engineering analysis?
No. It provides planning-level screening, prioritization and hourly inputs. Asset-specific thermal, voltage, protection and reliability analysis remains an engineering function.
See how EV ownership becomes an hourly utility load forecast.
Request a guided demonstration of customer probabilities, charging profiles, whole-home loads and current GIM applications.