Load Forecasting

Training for States on Distribution System and Distributed Energy Resources Planning

Presented by Julieta Giraldez, Kevala November 29, 2023

Load Forecasting – What Is the Status Quo?

• Demand has been flat for the Electricity end use in the United States from 1975 to 2022

past 20 years
• Utilities had time to “react” to
local load growth from new
customers and businesses
• Past consumption was a good
representation of future
consumption
Source: © Statista 2023

2
Load Forecasting – What Has Changed?

• Econometric modeling using historical Gross load versus net load example
data (typically load, weather) is not
sufficient to forecast future load
• Customers are adopting new
technologies behind-the-meter
• Need to understand gross load versus
net load
• Need to understand where and when
technologies are being adopted today
and in the future
• Rapid DER adoption trends are very
different than a new development or
business customer
• Past weather is not representative of
future weather
Source: © CAISO

3
Policy Is Greatly Influencing Load Forecasting
• DER adoption is heavily influenced by
federal/state/local/utility policies and goals
• Harder to quantify implications and what is
possible
• Initiatives and programs have to be converted
into quantifiable input assumptions on
technology adoption, utilization, operation

Source: National Public Utilities Council

Source: National League of Cities

4
Policy Is Greatly Influencing Load Forecasting

• Need to plan for longer time horizons

• Distribution planning has typically looked 3-5 years ahead
• Long lead time on grid assets and transmission constraints are
increasing the pressure on distribution planning
• Need to consider multiple scenarios

Distribution Planning

5
Past - Current - Future in Load Forecasting for Distribution Planning

● 5, 10, 30 year time horizon ● Customer level

● 5 year time horizon ● 8760 trends ● 8760 + disaggregated load
● Econometrics trends ● DER adoption and spatial components
● DER adjustments allocation ● Parametric distributions
● System level ● Substation / feeder level for variable to consider
● Deterministic ● Deterministic scenarios uncertainty
● Probabilistic

Source: Eversource
6
Who Performs Load Forecasting?

Forecasters Use Cases

Research organizations National and state studies
State energy offices Integrated resource planning, renewable
portfolio standard plan
Independent system operators
Resource adequacy
Utilities – load forecasting departments
(typically, the rates department) Transmission planning
Utilities – distribution planning department Rate design
Corporate forecast & revenue projections
Procurement
Distribution planning

Distribution planning has traditionally not used the

forecast from the load forecasting department.
7
Peak versus Energy Load Forecasting

• Load forecasting departments at utilities typically forecast energy and demand

separately
• Distribution Planning has traditionally only been concerned about substation/
feeder peak load to determine how big the infrastructure needs to be

Source: We Energies
8
Peak Load Forecast Modeling in Distribution Planning

• Historical peaks from SCADA

measurements at substation
and/or feeder-head Source: Kevala

• SCADA needs to be processed Example of Outliers for Abnormal Reconfiguration Event

to confirm the ”normal” peak
(vs. an abnormality)
• Typically, a manual and
burdensome task
• Generate a 1-in-10 (90th
percentile) load forecast based
on historical weather

Source: NREL

9
Load Forecast Modeling in Distribution Planning
Annual 90th percentile peak
Monthly 90th percentile peak

Source: Modified from ISO-NE

• Distribution Planning typically uses annual peak 1-in-10-year load forecasting at the
substation and/or feeder levels and might or not disaggregate top-down forecasts
for load or DERs
• New local large customer interconnection requests are added to the historical peak
10
Distribution Planning Load Forecasting

Source: NREL/Kevala
11
 Use-Case: Capacity Planning
“Long-Term” Capacity Planning (5-10
years): thermal evaluation at the
substation or feeder-head level.

Source: Kevala

12
Load Forecast - Key Input to Capacity Planning

● Spreadsheet exercise to predict peak

load at every substation and/or feeder
● Single deterministic forecast
● Overload criteria typically 100%
o When equipment is overloaded, it may fail

Source: PG&E

13
New Business Customers Driving Investments Is Reactive

● Load growth is consistently

missed

Source: Distribution Investment Deferral Framework:

Evaluation and Recommendations
14
Investments Consistently Needed

Source: Distribution Investment Deferral Framework:

Evaluation and Recommendations
15
“Allocation” or “Disaggregation” Using Load Shapes

● Increasing need to understand full load-shape profile to model

future peak load quantity and time of year and day
○ Overall load can be taken apart (disaggregated) to identify trends
in individual end uses
○ Customer segment at the substation/feeder level by customer
class is used for DER adoption and forecasts
■ Customer research department, advanced metering
infrastructure (AMI) data
● Full bottom-up models leveraging AMI and SCADA are starting to
be used
○ Kevala - CPUC Electrification Impacts Study - Part 1

See NREL/Berkeley Lab data, modeling tools, and reports on End-Use Load Profiles
16
“Allocation” or “Disaggregation” of Load and DER

● Load and DER forecasts

are often performed at Substation: Lat/Long

the zip code level Transformer Bank: La t/Lo n g, GNA lo a d in g, GNA ra tin g, SCADA MW

● Disaggregation step
from zip code to grid Feeder: La t/Lo n g, GNA lo a d in g, GNA ra tin g, SCADA MW

infrastructure (e.g., Service Transformer: La t/Lon g, KVA ra tin g

feeder) induces errors Parcel: Ce n su s, zip co d e , size

Premise: La t/Lo n g, kWh , ra te , DER

Source: Kevala

17
 Challenge to “Allocate” DERs - “Peanut Butter Spread”

● If DERs are allocated to substations/feeders

there is a risk of not capturing local peak
coincidence
● If DERs are not modeled using 8760, impact
to peak load could be over/under estimated

Source: PG&E
18
DERs Are Challenging the Peak Load Forecast Model

• Load forecast is now driven by customer adoption of DERs

• To understand the impact of DERs we need timeseries modeling
• Need to align top-down targets with distribution needs
• Load forecast is now driven by extreme weather events

Source: Kevala
19
 CPUC Electrification Impacts Study - Part 1 - Impact of EV Charging

Adding between 3.2M and 10.0M light-duty (LD) ZEVs by 2035 across the three IOUs has roughly the
same energy impacts as adding 2.9M to 8.7M residential customers’ worth of new energy demands.

Base Case High Electrification Accelerated High Electrification

ZEV adoption sources: ZEV adoption sources: ZEV adoption sources:
– LD: CEC 2021 IEPR Base Case – LD: CARB ACC II – LD: CEC 2021 IEPR Bookend Case
– Medium duty/heavy duty (MD/HD): CEC – MD/HD: CARB 2020 SSS (ACT & ACF) – MD/HD: CEC 2021 IEPR High Case
2021 IEPR Base Case
2035 ZEV-equivalent energy:
– 3.2M LDs: 2.9M residential customers 2035 ZEV-equivalent energy: 2035 ZEV-equivalent energy:
– 227k MD/HDs: 173k commercial customers – 10.0M LDs: 8.7M residential customers – 9.5M LDs: 8.2M residential customers
– 219k MD/HDs: 198k commercial – 231k MD/HDs: 164k commercial customers
customers

Three IOUs’ Total EV Energy (GWh) Three IOUs’ Total EV Energy (GWh) Three IOUs’ Total EV Energy (GWh)
Base Case High Electrification Accelerated High Electrification

Source: Kevala
20
Example Electrification Scenarios – Base Case versus High

Source: Kevala
21
DERs – Demand-Side Modifiers

• How to predict where (which substation and feeder) and

when will each technology be adopted?

Behind -the - Electric Vehicles (EV) and

Behind -the - Meter Building Electrification (BE)
Meter Battery Energy Electric Vehicle Service Equipment
Photovoltaics (PV)
Storage System (BESS) (EVSE)

Energy Efficiency (EE) Demand Response (DR) Pricing & Programs Smart Controls
(P&P)

Source: Kevala
22
DER Modeling Basis

• Each DER requires an adoption propensity model

Source: Kevala
23
Challenges with EE & BE Adoption and Behavior in Distribution Planning

● EE methods in distribution planning often rely on ratio of savings rather than

specific measure adoption
○ In contrast, for other DERs, the specific technology adopted is estimated along with
load implications (size and behavior) of that technology
○ The type of load conversion could dramatically impact the behavior and level of BE
adoption.
● Assumes uniform savings across baseline loads, potentially attributing savings
in hours when savings may not occur
○ For example, savings of 2% could be due primarily to lighting, yet lighting savings are
limited during the day or early mornings
○ Could miss compounding benefits from temperature-sensitive measures
○ Converting heating loads from natural gas to electricity (for both commercial and
residential sectors) could transition a customer with low energy use to a much higher
electric bill in exchange for a much lower (or nonexistent) gas bill
● Methods typically model savings proportional to size of customer’s load
○ While intuitive (customers with high energy usage potentially have more opportunities
for greater savings), this results in very large customers capturing the ‘target’ savings
first, potentially missing smaller premises that also could adopt
24
Challenges with EE & BE in Adoption and New Load Growth

Need to consider recent state and federal level legislation:

● IRA appliance rebates
● CA example
○ SB 1477 (2018) calls on the CPUC to develop two programs (BUILD and TECH) aimed at
reducing greenhouse gas emissions associated with buildings.
○ AB 3232 (2018) directs the California Energy Commission (CEC) to “assess the potential … to
reduce the emissions of greenhouse gases in … residential and commercial building stock by at
least 40 percent below 1990 levels by January 1, 2030.”
○ SB 68 (2021) directed the CEC to develop guidance and best practices to overcome barriers to
building electrification and electric vehicle charging equipment.
○ CEC 2022 building code - Encourages electric heat pump technology and electric-ready
requirements for other technologies for new construction

25
Deterministic Scenarios vs. Probabilistic Load Forecast

● Deterministic Scenarios
○ Change assumptions for final target and speed of DER adoption
○ Results in a range but does not quantify uncertainty
● Probabilistic Load Forecasting
○ Determines a range and probability distribution for each of the driving
variables of the forecast
○ Individual components of the load and DER forecast are turned into
probabilistic forecasts with calculated uncertainty
Challenge: How to combine uncertainty from every load and DER
model into one capacity planning model that can be used to make
investment decisions

26
Deterministic Scenario Matrix Design

Source: Kevala
27
 Probabilistic Load Forecasting

● Quantifies uncertainty for each scenario

○ Probabilistic component forecasts

Source: Appl. Sci. 2023, 13(11), 6520; https://doi.org/10.3390/app13116520

28
 Key Gaps and Needs in Distribution Planning Load Forecasting

• Statistical load
forecasting based on
historical load and
weather events will
miss extreme
weather events

30
Key Gaps and Needs in Distribution Planning Load Forecasting

• Statistical load forecasting based on historical load and weather

events will miss extreme weather events
• Hourly climate model projections are currently being developed

Source: EPRI
31
 Existing Load Forecasting for
Capacity Planning
● Capacity planning mismatch with
long-term changing policy goals
● Historical trends (load, weather, etc.)
are used to predict the future
● Allocation/forecasts not aligned with
electrical infrastructure and meters

Source: NREL/Kevala
32
Future Load Forecasting for Capacity Planning

● High spatial and temporal resolution for

load and DER forecasting
● Longer term forecast
● Scenario and probabilistic methods
● Include climate change models and extreme
weather events

Source: NREL/Kevala
33
Questions to Ask

• Does distribution planning coordinate with or take inputs from

the load forecasting department?
• Do you forecast peak load or some form of timeseries?
• What DERs are explicitly forecasted and modeled in your
distribution planning forecast?
• What weather data is used in your distribution planning load
forecast? Does it include the effects of climate change?
• Do you perform a single point load forecast, or do you consider a
range of scenarios and probabilistic methods to determine
infrastructure needs?

34
Questions?

[email protected]

35