Residential electricity demand on CAISO Flex Alert days: a case study of voluntary emergency demand response programs

Abstract

The California Independent System Operator (CAISO) utilizes a system-wide, voluntary demand response tool, called the Flex Alert program, designed to reduce energy usage during peak hours, particularly on hot summer afternoons when surges in electricity demand threaten to exceed available generation resources. However, the few analyses on the efficacy of CAISO Flex Alerts have produced inconsistent results and do not investigate how participation varies across sectors, regions, population demographics, or time. Evaluating the efficacy of DR tools is difficult as there is no ground truth in terms of what demand would have been in the absence of the DR event. Thus, we first define two metrics that to evaluate how responsive customers were to Flex Alerts, including the Flex Period Response, which estimates how much demand was shifted away from the Flex Alert period, and the Ramping Response, which estimates changes in demand during the first hour of the Flex Alert period. We then analyze the hourly load response of the residential sector, based on ~200,000 unique homes, on 17 Flex Alert days during the period spanning 2015 to 2020 across the Southern California Edison (SCE) utility’s territory and compare it to total SCE load. We find that the Flex Period Response varied across Flex Alert days for both the residential (-18% to +3%) and total SCE load (-7% to +4%) and is more dependent on but less correlated with temperature for the residential load than total SCE load. We also find that responsiveness varied across subpopulations (e.g., high-income, high-demand customers are more responsive) and census tracts, implying that some households have more load flexibility during Flex Alerts than others. The variability in customer engagement suggests that customer participation in this type of program is not reliable, particularly on extreme heat days, highlighting a shortcoming in unincentivized, voluntary DR programs.

Full text

Environ. Res.: Energy 1(2024) 015002 https://doi.org/10.1088/2753-3751/ad0fda
OPEN ACCESS
RECEIVED
18 July 2023
REVISED
8 November 2023
ACC EPT ED FOR PUB LICATI ON
23 November 2023
PUBLISHED
8 December 2023
Original content from
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence.
Any further distribution
of this work must
maintain attribution to
the author(s) and the title
of the work, journal
citation and DOI.
PAPER
Residential electricity demand on CAISO Flex Alert days: a case
study of voluntary emergency demand response programs
McKenna Peplinski∗and Kelly T Sanders∗
Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, CA, United States of America
∗Authors to whom any correspondence should be addressed.
E-mail: peplinsk@usc.edu and ktsanders@usc.edu
Keywords: CAISO, Flex Alert, residential electricity, smart meter
Supplementary material for this article is available online
Abstract
The California Independent System Operator (CAISO) utilizes a system-wide, voluntary demand
response (DR) tool, called the Flex Alert program, designed to reduce energy usage during peak
hours, particularly on hot summer afternoons when surges in electricity demand threaten to
exceed available generation resources. However, the few analyses on the efficacy of CAISO Flex
Alerts have produced inconsistent results and do not investigate how participation varies across
sectors, regions, population demographics, or time. Evaluating the efficacy of DR tools is difficult
as there is no ground truth in terms of what demand would have been in the absence of the DR
event. Thus, we first define two metrics that to evaluate how responsive customers were to Flex
Alerts, including the Flex Period Response, which estimates how much demand was shifted away
from the Flex Alert period, and the Ramping Response, which estimates changes in demand during
the first hour of the Flex Alert period. We then analyze the hourly load response of the residential
sector, based on ∼200 000 unique homes, on 17 Flex Alert days during the period spanning
2015–2020 across the Southern California Edison (SCE) utility’s territory and compare it to total
SCE load. We find that the Flex Period Response varied across Flex Alert days for both the
residential (−18% to +3%) and total SCE load (−7% to +4%) and is more dependent on but less
correlated with temperature for the residential load than total SCE load. We also find that
responsiveness varied across subpopulations (e.g. high-income, high-demand customers are more
responsive) and census tracts, implying that some households have more load flexibility during
Flex Alerts than others. The variability in customer engagement suggests that customer
participation in this type of program is not reliable, particularly on extreme heat days, highlighting
a shortcoming in unincentivized, voluntary DR programs.
1. Introduction
In California, the state’s largest balancing authority, California Independent System Operator or (CAISO),
utilizes a voluntary demand response (DR) tool known as Flex Alerts. When a Flex Alert is issued, electricity
consumers are asked to voluntarily reduce their electricity usage for a specified period in time to reduce
strain on the grid during periods when grid reliability is threatened [1]. The Flex Alert program has become
an important tool for managing California’s electric grid, which has been challenged by an increasing
frequency and intensity of extreme heat events, in conjunction with a rapid increase in variable renewable
energy, which provide challenges for reliable grid operation.
Flex Alerts have helped avoid rolling blackouts by reducing total system load during times of grid stress
particularly on hot summer afternoons when air-conditioning (AC) use is high [1,2]. During heat waves,
electricity demand often surges as customers turn on AC to stay cool [3]. To avoid disruption, grid operators
must secure enough generation resources to meet rising demand, at the same time that electricity generators
themselves might experience reduced capacity because of the high temperatures. For example, extreme heat
© 2023 The Author(s). Published by IOP Publishing Ltd

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
can reduce reliable thermal power plant operation via lower power plant efficiencies [4], or in extreme cases,
due to inadequate or disrupted cooling water resources [5]. The efficiency of solar photovoltaic panels are
also reduced on hot days [6]. When extreme heat occurs during periods of drought, low hydropower
resources can further exacerbate reductions in generation capacity [7]. Heat may also cause losses in
transmission lines through both lowered line ratings and a need to de-energize in instances of wildfire risk
[8,9]. Extreme heat events in the past have forced utilities to resort to involuntary load shedding (i.e. rolling
blackouts) when available power generation was inadequate to meet demand [10–12]. While rolling
blackouts can effectively stabilize the power grid, they pose significant public health risks and have been
shown to more adversely affect minorities and populations with lower socioeconomic status [13].
In addition to extreme heat, high penetrations of variable renewable energy, namely solar, on California’s
electric grid have prompted a set of emerging challenges that can threaten supply–demand balancing,
particularly during periods of peak demand. These challenges are especially acute on high electricity demand
days in the early evenings when solar PV generation falls at the same time that net demand (i.e. total demand
less total variable renewable generation) approaches its peak. To accommodate the rapid increase in net
demand, electricity generators from other (typically more dirty) resources have to quickly ramp up their
generation, which can threaten grid reliability as there are physical constraints on the rate at which fast
generators can be dialed up [14]. Conversely, in mid-day periods when solar resources are high and net load
is at a low, CAISO has been challenged by solar overgeneration (i.e. periods when solar generation exceeds
what the grid can accommodate), typically on sunny Spring days when hydropower and/or wind resources
are also high [15]. During these periods system operators often curtail this load to avoid damage to the grid
[16].
Flex Alerts can alleviate some of the stress of fossil fuel power generator ramping and solar
overgeneration by encouraging customers to shift electricity demand from on-peak to off-peak hours. In
CAISO, shifting load to off-peak hours when solar and wind output is high has the added benefit of
increasing the amount of load met with emissions-free renewable generation. For some loads, this temporal
shifting might cause a net increase in daily electricity usage, for example if customers precool their homes by
running their air-conditioners more intensely in early afternoon in efforts to relieve cooling during on-peak
hours [17]. (However, since electricity is cleaner and cheaper prior to Flex Alerts, there are likely indirect
emissions and cost benefits in addition to grid reliability benefits, even in these cases [18]).
Most analyses of CAISO Flex Alerts have focused on impacts to systemwide or regional demand, and
therefore, do not capture how participation in each DR event may have varied across sectors, spatial and
temporal extents, or population demographics. CAISO itself has stated that there have been significant drops
in overall demand during typical ramping hours on Flex Alert days, suggesting that they are a useful tool for
shedding load [2,19,20]. However, an energy consulting firm released a load impact evaluation of
California’s Flex Alert program in 2014 using PG&E load data from three different Flex Alert days in 2013,
roughly 10 years after program deployment, and did not find a statistically significant difference in the load,
compared to reference days [21]. In recent years, there have been calls by the California Public Utilities
Commission (CPUC) and investor-owned utilities (IOUs) to study consumer awareness of the program [22],
but there has very little analysis on the efficacy of or participation in CAISO’s Flex Alert program, in part due
to the lack of high-resolution, publicly available data at a regional scale. As a result, our understanding of
how different populations of electricity consumers respond when Flex Alerts are issued is limited.
In this research we aim to both evaluate the effectiveness of Flex Alerts as well as define what it means for
a Flex Alert to be ‘effective’, as there exists no standard metric in the literature. We use five years of hourly
smart meter electricity data from approximately 200 000 homes in Southern California to analyze the energy
demand of residential customers on Flex Alert days. The following research questions are addressed: (1) how
effective have Flex Alerts been in reducing pressure on generation fleet ramping (‘Ramping Response’) and
residential sector demand (‘Flex Period Response’) during Flex Alert hours? (2) do factors including daily
maximum temperature, weekend–weekday scheduling, and frequency of Flex Alert issuance affect Flex
Period Responses on Flex Alert Days?, and (3) what subpopulations of customers are most likely to change
their behavior during Flex Alert Periods? the results of this study will provide insight into the efficacy of
voluntary DR programs and how they can be tailored to better engage and motivate different groups of
residential customers.
2. Background
While few studies have analyzed the CAISO Flex Alert campaign specifically, many studies have attempted to
quantify the full potential of residential DR by simulating or modeling energy behavior and customer
response to certain signals [23–26]. These studies focus on how ‘flexible’ certain loads could be (i.e. how
much of demand could potentially be shifted to other hours), but the results of these analyses do not give
2

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
insight into how actual electricity users behave when DR events are called. Energy behavior, especially in the
residential sector, is driven by a multitude of factors and is highly variable and unpredictable [27]. In the case
of DR programs, users will not always respond to price signals or incentives consistently or in the most
economical way [28]. A review of residential engagement in DR trials and programs found that modeling
studies generally have optimistic assumptions about consumer engagement that are not realized, meaning
the real-world customer response is lower than estimates in the peer-reviewed literature [29].
To gain better insight into how electricity users and different subpopulations will actually respond to DR,
many studies analyze the load data of customers who participated in DR pilot programs and trials [30–34].
For example, one study evaluated the response of 483 residential customers to critical-peak pricing in
California, finding that users with higher electricity demand and users in cooler climate zones were more
responsive to critical-peak pricing and reduced their demand by larger percentages, compared to users with
smaller loads and in warmer locations [30]. However, the number of participants enrolled in these DR pilots
are not typically representative of an entire region (e.g. a couple hundred to a couple thousand homes
[35–38]), which limits insight into the overall efficacy of the program or how responses differ within a
region. Further, they focus on pricing or incentive driven programs [34,39,40] rather than programs like
Flex Alerts where customers are called on to voluntarily reduce demand without incentive [41,42].
Voluntary DR programs that offer no financial incentive for customers who shift or shed their load, such
as Flex Alerts, are far less common in practice and in the literature. However, it has been shown that
customers can be persuaded to conserve energy without incentive when emergency situations occur. In 2009
when a transmission line was severed in Juneau, Alaska, the town’s residents responded to conservation
requests and decreased their electricity use by 30% compared to the same time in the previous year [42].
Similarly, California developed a statewide public information campaign, titled Flex Your Power, during its
2001 energy crisis, that urged residents and businesses to reduce both their overall and peak demand. As a
result of the campaign, additional DR programs, and efficiency measures, the peak electricity demand was
reduced by an estimated 6,369 MW, with 2,616 MW reduction being credited to voluntary conservation
alone [43]. These examples indicate that voluntary DR programs can be an effective tool to help avert energy
emergencies, but to replicate their success, further analyses of the factors that impact the level of response are
needed.
Evaluating the efficacy of DR programs is difficult because there is no precise ground truth of what
energy demand would have been in the absence of the DR tool since variables such as temperature and other
meteorological conditions, day of the week, holidays, etc. influence day-to-day and hour-to-hour demand.
Most studies establish a reference case (e.g. predicted demand [30], demand from similar, non-participating
customers [34], or a constructed baseline determined from demand during a reference period [32,33,39,42,
44]) to serve as a proxy that can be compared to the observed demand to estimate how successful a DR
program was. However, a more robust, standardized analytical method that assesses the extent to which
customers engaged in DR events would be useful to systematically quantify and compare the efficacy of DR
programs across time, space, and customer demographics and the role DR can play in achieving system
reliability.
3. Methods
3.1. Datasets
Flex Alert data were retrieved from CAISO’s Grid Emergencies History Report for the years 2015–2016 and
2018–2020 [45]. (These time periods were selected to align to our available hourly household-level electricity
dataset described below). Data regarding the CAISO Flex Alert issuance date, targeted period (e.g.
4pm–9pm), and regional coverage (e.g. all of CAISO vs. Southern California) were extracted. In total, 18 Flex
Alerts were issued during the study period with regional coverage that included Southern California. We
chose to omit one Flex Alert (20 June 2016) from the study because the length of the alert (issued from 10 am
to 9 pm) was markedly longer than others. We kept another Flex Alert, issued on 7 September 2020, in the
analysis, although it occurred on a major holiday (Labor Day), which we acknowledge may have impacted
customer response level.
Residential sector data comprised of hourly-smart meter electricity records for roughly 200 000 unique
homes across the entire Southern California Edison (SCE) Investor-Owned Utility (IOU) service area were
selected to be statistically representative of the SCE region with 99% certainty. SCE provided the smart-meter
dataset with matching street-level addresses for the years 2015–2016 and 2018–2020. Data were stored on a
high security data account to abide by the privacy requirements of the SCE. These household-level data were
analyzed to evaluate how responsive the residential sector and subpopulations within the residential sector
3

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
were to Flex Alerts. In this analysis, we use the normalized shape of our hourly smart meter dataset (i.e. after
aggregating customer load at each hourly increment) as a proxy for SCE’s normalized residential sector load
and refer to it as the ‘residential load’.
Total hourly load data for SCE, which encompasses all end-use sectors, on each Flex Alert day were
retrieved from CAISO as a reference to compare to the residential smart meter data (referred to as ‘total SCE
load’).
Local and regional temperatures were used in this analysis to investigate the impact of temperature on
customer response during Flex Alerts and to identify similarly hot days to use as reference. Three sources of
land-based weather station networks were used in this study: CIMIS, EPA AQS, NOAA. In total, historical
dry bulb temperature records were retrieved from 112 different stations throughout the study region
[46–48], and each household was matched to the nearest weather station. Daily average and maximum
temperatures were computed on Flex Alert days for every census tract in the study area, as well as for the
study area as a whole, by taking the average of each household’s assigned temperature in the area of interest.
The temperatures were used to compare weather conditions across Flex Alert days and determine
comparable days (described in the following section). Daily temperatures were assumed to be equal for the
residential load and total SCE load analyses. Each house was also matched to its corresponding California
climate zone, as defined by the California Energy Commission, to gain insight into how consumer response
varied across micro climates [49].
Census tract level data from CalEnviroScreen 3.0 and the U.S. Census Bureau were retrieved to compare
energy consumption behavior across different subpopulations [50,51]. These data include poverty
percentile, education percentile (e.g. the percent of households in a census tract where a resident has at least a
high school education—shown in SI), and income. Each household is characterized by the census tract it is
situated in, as household level socioeconomic data is not available.
3.2. Response metrics
To define how ‘responsive’ a household was to the Flex Alerts, it was prudent to define a reference scenario to
estimate what hourly electricity usage behavior might have been in the absence of the Flex Alert. Since it is
impossible to know what actual electricity use behavior would have been in the absence of a Flex alert on a
Flex Alert day, we assigned three ‘comparable days’ to each of the 17 Flex Alerts studied. The set of
comparable days were selected based on 4 criteria, including that they (1) occurred in the two weeks leading
up to or two weeks following the given Flex Alert, (2) shared the same ‘day type’ (i.e. weekday or weekend) as
the Flex Alert, (3) were not classified as a Flex Alert day, and (4) represented the three hottest days based on
the region’s daily maximum temperature. Every census tract was also assigned three comparable days for
each Flex Alert day using the census tract’s daily maximum temperature. Thus, the assigned comparable days
may differ across census tracts but are always the same for every household within a census tract.
The electricity demand of each unique customer ion each of the three comparable days, E(C1)
i,h,E(C2)
i,h, and
E(C3)
i,h, respectively, is averaged to generate an estimate of the electricity demand of each customer in hour h
on the Flex Alert day had the Flex Alert not been issued. We refer to this as the reference electricity demand,
E(R)i,h described in equation (1). We use this metric as a proxy to estimate hourly demand across the
reference day in equations (2)–(4)
E(R)
i,h=
E(C1)
i,h+E(C2)
i,h+E(C3)
i,h
3.(1)
Two metrics were developed to evaluate how responsive customers were to each Flex Alert. The metrics
were calculated for the SCE’s entire service territory and at the census tract level to explore how the response
varied across spatial extents and subpopulations. (Note: In equations (1)–(5) no summation across
customers iis required when calculating the Flex Period Response and Ramping Response of SCE’s total load
data as these data are already aggregated into hourly values representing all customers and all sectors in SCE
territory).
The first metric, described by equation (2), is referred to as the Flex Period Response and is a proxy to
estimate the change in the percent of total daily demand that took place during targeted Flex Alert hours on
the Flex Alert day, F, compared to the same day if a Flex Alert was not issued (i.e. the reference day, R).
Hence, to calculate the Flex Period Response, we calculate, P(F), representing the percentage of total daily
demand that occurred within the Flex Alert period by summing the total amount of electricity demand by
customers across the spatial extent of interest (i.e. census tract or total SCE region) between hour s, defined
to be the start of the Flex Alert period, and hour e, the end of the Flex Alert period, and divide by the total
4

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
daily electricity usage of all customers in the dataset on the Flex Alert day, as shown in equation (2). Then we
repeat the calculation in equation (3) to calculate the same value for the percentage of demand occurring
within the Flex Alert hours on the reference day referred here to as P(R)
P(F)=∑i∑e
h=sE(F)
i,h
∑i∑23
h=0E(F)
i,h
×100.(2)
P(R)=∑i∑e
h=sE(R)
i,h
∑i∑23
h=0E(R)
i,h
×100.(3)
The Flex Period Response (units of percent change) represents the difference between, P(F)and P(R)as
shown in equation (4)
FlexPeriod Response =P(F)−P(R)
.(4)
The second metric, described as the Ramping Response (units of percent change) in equation (5), is a
proxy to estimate the difference in how demand changed across the first hour of the Flex Alert Period (i.e. the
hour spanning h=sto h=s
+
1) on the actual Flex Alert day versus the reference day
RampingResponse =(∑iE(F)
i,h=s+1−∑iE(F)
i,h=s
∑iE(F)
i,h=s
−∑iE(R)
i,h=s+1−∑iE(R)
i,h=s
∑iE(R)
i,h=s)×100.(5)
The Flex Alert period varied across Flex Alert days so the targeted period in equation (2) through (4)
reflects the period defined by the unique Flex Alert listed in CAISO’s Grid Emergencies History Report [45].
For example, on June 30th, 2015 the alert was issued to begin at 2 pm and end at 9 pm (i.e. s=2pm to
e=9pm) and on 14 August 2020, the alert was issued to begin at 3 pm and end at 10 pm (i.e. s=3pm to
e=10pm). For each set of comparable days, the sand evalues are defined based on the corresponding Flex
Alert day.
4. Results and discussion
The Flex Period and Ramping Responses of both the residential load and the total SCE load were calculated
to understand the varying success of the program and identify factors that might have motivated customer
engagement. In figure 1, the Flex Period Response of the (a) residential load and (b) total SCE load on each
Flex Alert day is plotted against the study region’s daily maximum temperature on the day of the Flex Alert.
In figure 1(c), a timeline of the Flex Alerts across the five years of data is shown with the corresponding Flex
Period Response on those days (note: desired Flex Period Responses are negative, representing reductions in
load). The shape of the points on the scatter plot represents whether the date was a weekday or weekend, and
the shade of each point refers to the order within each year that the Flex Alerts were issued. (We were
interested in understanding the frequency and timing of Flex Alerts throughout the year to see if there might
be fatigue across the customer base when many Flex Alerts were issued within a short duration of time).
Points that are outlined in red represent Flex Alert days that were hotter in temperature than all three of their
comparable days.
The scatter plots show that there is a correlation between the study region’s Flex Period Responses and
the region’s daily maximum temperature, and the slope and r value of the line of best fit can give insight into
how dependent and correlated the Flex Period Responses are to daily maximum temperature. In general, the
residential load and total SCE load were more likely to reduce demand during Flex Alert hours on days with
relatively cooler temperatures (i.e. higher magnitude, negative Flex Period Response). When compared to
total SCE load, the Flex Period Response of the residential load is more strongly dependent on (slope of 0.52
versus 0.34) but less correlated to (rvalue of 0.58 vs 0.69) the daily maximum temperature. No observable
trends were noted across weekday/weekends, the number of Flex Alert issued in the year, or whether the Flex
Alert was the hottest day of the comparable days. Additionally, the timeline in figure 1does not show any
trend of consumers being less responsive on consecutive days of issued Flex Alerts (i.e. there was no response
fatigue after multiple consecutive days of Flex Alerts).
Table 1summarizes Flex Alert days and the responses of both the residential sector and all SCE customers.
Compared to the Flex Period Response, the Ramping Response was much less strongly dependent on the
region’s daily maximum temperature, with a slope of 0.07 and 0.05 for the residential load and the total SCE
load. On average, the Ramping Response also has a lower magnitude, negative value for both the residential
SCE and total SCE load. It is important to note that the shape of the load profile on the three comparable
5

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
Figure 1. Flex Period Response of the (a) the residential SCE load and (b) total SCE load on each Flex Alert day issued in
2015–2016 and 2018–2020 versus the region’s daily maximum air temperature. (c) A timeline of when each Flex Alert was issued
with the corresponding Flex Period Response.
days strongly impacts the Ramping Response, and if the demand peaks earlier or later than is typical it could
lead to an under or overestimation of the responsiveness. For example, on 11 June 2019, the demand on one
of the three comparable days is already declining before the start of the Flex Alert period, which increases the
value of the computed Ramping Response (i.e. positive or lower magnitude, negative value). Because there is
no ground truth to compare against, this is a difficult limitation to avoid. (Note: the hourly load profiles of
the residential load and total SCE load on all of the Flex Alert days are available in the SI).
Table 1also highlights that the overall Flex Period Response of both the residential and total SCE load is
inconsistent, and certain Flex Alert days appear to be effective (i.e. a negative Flex Period Response) while
others are not. On the most effective Flex Alert days (apart from the Flex Alert that fell on a major holiday),
the Flex Period Responses of the residential loads are −11% (30 June 2015 and 1 July 2015), but most days
have more modest values, or even positive Flex Period Response values, indicating that load actually
increased during the targeted period when compared to the estimated reference day. The average Flex Period
Response of the residential and total SCE load are −4% and −2%, respectively. It is likely that large industrial
and commercial loads are already participating in existing SCE DR programs (e.g. [52]) that incentivize
them to shift or shed their large loads (and possibly on selected comparable days, in addition to Flex Alert
days), which might partially explain why these customers have a lower average Flex Period Response and less
sensitivity to temperature.
While the days with positive Flex Period Reponses may seem like days in which the program was
ineffective, our reference day demand is only an estimate of what demand would have been in the absence of
a Flex Alert, so we have no precise ground truth for comparison. Thus, it is possible that customers still used
less demand during the Flex Alert period on these days than they would have in the absence of a Flex Alert,
particularly on days where the Flex Alert day was hotter than the three comparable days used to calculate the
reference.
As observed in figure 1, the Flex Period Response of both the residential and total SCE load varies across
the Flex Alert days. Figures 2(a) and (b) highlights the load profiles of the residential load and total SCE load
on two different Flex Alert days (30 June 2015 and 1 July 2015), which had the largest flex period responses in
the sample of Flex Alerts studied (both −11%). The load profiles of the comparable days, used to calculate
the reference day hourly load, are also shown for context. As a note, the three comparable days exhibited
strong consistency in the shape of their load profiles for a corresponding Flex Alert day (refer to the SI for the
6

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
Table 1. Summary of CAISO’s Flex Alerts from 2015–2016 and 2018–2020 with the corresponding Flex Period Response and Ramping
Response of SCE’s residential load and total load.
Residential SCE Load Total SCE Load
Date Day of week
Daily max
temp (F)
Ramping
Response
Flex Period
Response
Ramping
Response
Flex Period
Response
30 June 2015 Tuesday 90.4 −9% −11% 0% −5%
1 July 2015 Wednesday 86.5 −1% −11% −2% −7%
27 July 2016 Wednesday 93.3 −2% −6% 0% −4%
28 July 2016 Thursday 92.2 −1% −5% 1% −4%
24 July 2018 Tuesday 96.3 1% −1% 2% 0%
25 July 2018 Wednesday 93.4 1% −2% 2% −1%
11 June 2019 Tuesday 92.5 0% 2% 1% 0%
14 August 2020 Friday 97.5 0% −1% 3% 0%
16 August 2020 Sunday 94.9 −3% −6% −2% −3%
17 August 2020 Monday 93.9 −1% −2% −1% −2%
18 August 2020 Tuesday 102 −2% −6% −1% −4%
19 August 2020 Wednesday 97.3 −1% −6% −1% −4%
5 September 2020 Saturday 106.2 −2% 3% 2% 4%
6 September 2020 Sunday 108.3 −3% −2% −1% 0%
7 September 2020aMonday 88 −7% −18% 0% −7%
1 October 2020 Thursday 99.7 −1% −1% 2% 0%
15 October 2020 Thursday 94.2 1% −2% 5% −1%
Average response −2% −4% 1% −2%
Median response −1% −2% 0% −2%
Slope of response metric to temperature metric 0.07 0.52 0.05 0.34
rvalue of response metric to temperature metric 0.11 0.58 0.16 0.69
aFlex Alert fell on a federal holiday.
Figure 2. (a), (b) Normalized hourly electricity load profile of residential load (purple) and total SCE load (red) on two different
Flex Alert days (solid lines) compared to the hourly load profiles on the comparable days (dashed lines). (c), (d) And hourly
percent change in electricity demand on two Flex Alert days (outlined in black) and their corresponding comparable days.
load profiles of the comparable days on all Flex Alert days). The Flex Period Response suggests that the DR
event was effective on both of these days; overall, customers used less of their daily load during Flex Alert
hours than they did on comparable days, reducing the generation resources needed during those periods.
While significant Flex Period Responses were observed during Flex Alert hours on both days in figure 2,
the shape of the responses differs. On 30 June 2015, the residential load sharply fell in the hours leading up to
and after the first hour of the Flex Alert and remained relatively low before peaking again in the final hours of
the Flex Alert. The heat maps in figures 2(c) and (d) underscore this behavior, illustrating a significant
decrease in demand occurring between hours 14 and 15, followed by an increase between hours 18 and 19,
which is markedly different from the patterns of changes in hourly demand on the three comparable days
occurring before or after the 30 June Flex Alert. In contrast, on 1 July 2015 there is no sharp drop in the
residential load; instead, the demand is consistently lower than the comparable days leading up to and
throughout the hours of the issued alert. However, on both Flex Alert days, there are ramping benefits
meaning that hourly increases in load during the initial hours of the Flex Alert are less than comparable days.
7

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
Figure 3. (a), (b) Choropleth map of the census tract level Flex Period Response of SCE’s residential load on two different Flex
Alert days. Areas in blue consumed a lower percentage of their total daily electricity demand during Flex Alert hours than they did
on reference days, while areas in red used a higher percentage of their total daily electricity demand during Flex Alert hours than
they did on reference days.
However, we observe slight penalties, in the case of the June Flex Alert, when demand spikes in the final
hours of the Flex Alert period.
The load profiles also highlight some differences in how the residential sector consumes energy
throughout the day versus all sectors. While both normalized load profiles show peaking behavior in the late
afternoon through evening hours, the residential load’s peak is significantly higher than the total SCE load’s
peak. Hence, the residential sector drives much of the peaking behavior that occurs during typical Flex Alert
hours, as much of the residential sector’s demand is driven by AC use. While it might be difficult for some
households to shift or shed their load when temperatures are hot, the high percentage of demand that takes
place during the hours of interest underscore the opportunity that residential consumers present for
participating in DR programs aimed to reduce peak electricity consumption.
The response to Flex Alerts across residential customers is not uniform. To observe variations across both
the microclimates of the study region and subpopulations, the residential Flex Period Response of each census
tract was mapped across SCE territory. The choropleth maps of the residential load’s Flex Period Response
on two different Flex Alert days are depicted in figure 3. Figure 3(a), represents 30 June 2015, a day where the
alert prompted a relatively large Flex Period Response of −11% for the residential load (compared to the
average residential load’s Flex Period Response across all Flex Alert days of −4%). On this day, the response
in the study region is relatively consistent with reductions in loads across most census tracts. Conversely, on
11 June 2019, the average Flex Period Response for residential customers across SCE’s service region was
+2%, meaning that there was an average increase in load during the Flex Alert compared to electricity
consuming behavior on similar days. However, when the responses are mapped at the census tract level, there
are many census tracts that were responsive to the Flex Alert (i.e. had a negative Flex Period Response value).
We also investigated how socioeconomic factors influence a customer’s participation in Flex Alerts.
Figures 4(a) and (b) depict the hourly residential load in MWh by electricity demand percentile and income
percentile, respectively, (binned according to total annual demand in each year) on a Flex Alert day, 30 June
2015. Figures 4(c) and (d) depict the percent of daily residential electricity load split across each hour of the
day by electricity demand percentile and income percentile, respectively, on the same Flex Alert day. From
the load profiles, we observe that higher income and higher demand customers have larger demand
reductions during Flex Alert hours than customers with lower income and demand. The heat maps in
figures 4(e) and (f) show percent change in hourly demand on the same Flex Alert day, again by demand and
income percentiles. Here we see that high income, high demand customers have a steeper decline in demand
during the initial hours of the flex alert than low income, low demand customers. These results can be
explained in part by previous studies on energy poverty, which have found differences in the energy behavior
of low-income and high-income utility customers. Low-income customers typically use significantly less
energy than their high-income counterparts and more consistently engage in energy limiting behaviors to
reduce costs [53]. Thus, when DR events occur, it is difficult for low-income customers to further reduce
their demand.
These results illustrate that average households in some census tracts likely have greater flexibility to
respond to Flex Alerts than in others. Despite the wide disparities in climate, housing stock, and population
demographics across and within both the census tracts and subpopulations studied, we can draw some
conclusions on some common characteristics of the households most likely to respond. For example,
customers with larger demand (who also tend to be higher income) are more likely to modify electricity
8

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
Figure 4. Hourly residential electricity load by (a) income percentile and (b) demand percentile and normalized hourly residential
electricity load by (c) income percentile and (d) demand percentile. Heat maps of hourly percent change in demand by (e)
income percentile and (f) demand percentile on a Flex Alert day, 30 June 2015. Note: the 10th percentile refers to the lowest
income and demand percentiles, and the 100th percentile refers to the highest income and demand percentiles.
consuming activities and might have access to technology that enables flexibility [54]. In contrast, a customer
that is already very energy conscious and/or has few discretionary electric loads (who also tend to be lower
income) might not be able to shift much of their load to other periods of the day. These results, on one hand,
indicate that the households with the largest potential to contribute large reductions in peak demand
(i.e. high consumers) are typically the households that are most likely to respond to Flex Alerts, which is
valuable for program success. However, on the other hand, our census tract results across the SCE region
indicate that responsiveness to Flex Alerts on extreme heat days (particularly in the hottest regions) tend to
be lower than Flex Alerts issued on comparatively cooler days (and in cooler regions). Hence, the Flex Alert
program might not be very effective on the most extreme heat days when grid resources tend to be most
strained.
5. Conclusion
The results of this study show that voluntary DR programs, even without financial or other incentives, can
significantly influence energy behavior, especially in the residential sector, where peak electricity usage tends
to be high compared to other sectors. On the Flex Alert days with the greatest percent reduction of daily
consumption during the Flex Alert hours, the residential load and total SCE load had Flex Period Responses
of −11% and −7%, respectively (excluding the Flex Alert that fell on a federal holiday). These values
represent meaningful decreases in the percent of daily demand used during Flex Alert hours suggesting that
Flex Alerts have provided a valuable tool to help maintain grid reliability on days when the grid has been
stressed. However, there are also days when responsiveness appears to have been somewhat negligible and
regional results suggest that Flex Alert responsiveness tends to be reduced on days and in locations
experiencing extreme heat (and electricity resources were presumably most strained).
Our investigation of differences across subpopulations and census tracts implies that some households
have more flexibility or ability to shift or shed their load and that this flexibility can vary significantly across
Flex Alert days. While we do not have full transparency into what drives these differences, which are likely
due to variety of factors (e.g. outdoor temperature, higher efficiency homes, more flexible schedules, more
discretionary loads during Flex Alert hours, and the health, wealth, and age of occupants, etc), we can
generate some broad insights. From our analysis we see clear overlaps between household income, total
electricity demand, and Flex Period Response; customers with large electricity demand often also have high
income, and hence, more demand flexibility. However, future analyses should focus on teasing out the other
factors that drive differences in customer engagement and literacy in the program across subpopulations to
gain a better understanding of what factors influence customer participation so that messaging for future
events can be tailored to increase engagement. Future work might also look into the efficacy of trying to drive
9

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
larger Flex Alert responses in cooler regions on extreme heat days, when responsiveness in the hottest regions
is likely to be muted.
This study underscores some of the shortcomings of unincentivized, voluntary DR programs, since it
difficult to predict how much additional capacity can be gained by issuing Flex Alerts, limiting their benefits
to grid operators for longer term planning. As the frequency and intensity of extreme heat events grow, a
viable, dynamic grid will likely require more flexibility than a program such as Flex Alerts can provide,
meaning robust, incentivized DR programs with reliable participation are needed to ensure resources are
available during periods of high demand. Most incentivized DR programs in California over the past decade
have been geared towards industrial and commercial customers with large loads for load shedding and load
shifting, but the results of this study show that there is huge potential for demand-side management
programs through the aggregation of smaller residential load reductions. Currently, DR programs are
decentralized and administered by either the state’s three IOUs, CPUC jurisdictional entities, or commercial
DR providers, and residential customers can elect to participate. However, the future of California’s electricity
systems will likely include a more formal approach to DR, such as universal opt-in access to real-time energy
and capacity pricing, as was outlined in the CPUC Energy Division’s 2022 ‘CalFUSE’ proposal [55]. (Note: In
May 2022, the California Public Utility Commission launched a pilot program administered by a customer’s
IOU, called ‘Power Saver Rewards’ that pays customers in bill credits when they respond to Flex Alerts [56].
That incentivized pilot program was not implemented during the period of study).
As markets for incentivized DR are developed, attention should be given to how to best optimize these
programs as there are conflating factors that can limit their overall success. For example, although wealthier
customers have the capability to load shift, studies have shown that they are harder to incentivize financially
[57]. And while lower income customers are more likely to respond to pricing signals, their participation
does not offer as much flexibility to the grid as higher income customers because their loads are generally
smaller [58,59]. Further, low-income, energy insecure customers may already be engaging in energy limiting
behaviors due to their own financial constraints. Thus, it is important to design programs that distribute the
benefits to vulnerable households, whose ability to participate is limited [60], and hence, will not benefit
equally from financial incentives or lowered electricity prices in comparison to wealthier, higher consuming
users [61]. Still, regardless of who participates, successful DR programs that shift load to off peak hours
benefit the whole customer base through reductions in overall electricity costs that reflect a transfer of wealth
from electricity generators to electricity consumers [62,63].
Data availability statement
The data cannot be made publicly available upon publication because they contain sensitive personal
information. The data that support the findings of this study are available upon reasonable request from the
authors.
Acknowledgments
This work was funded in part by the National Science Foundation under Grants CBET-CAREER 1845931
and CBET-CAREER 1752522. Computation for the work described in this paper was supported by the
University of Southern California’s Center for Advanced Research Computing (carc.usc.edu). We also thank
Southern California Edison for access to the smart meter data.
Funding
The funding sources that supported this work include:
•CBET-CAREER 1845931
•CBET-CAREER 1752522
ORCID iDs
McKenna Peplinski https://orcid.org/0000-0002-6353-6211
Kelly T Sanders https://orcid.org/0000-0003-4466-0054
10

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
References
[1] California ISO What is a Flex Alert? (available at: www.flexalert.org/what-is-flex-alert)
[2] California ISO 2021 Californians do conserve when asked—Flex Alerts are vital (available at: www.caiso.com/about/Pages/Blog/
Posts/Californians-Do-Conserve-When-Asked-Flex-Alerts-Are-Vital.aspx)
[3] Auffhammer M, Baylis P and Hausman C H 2017 Climate change is projected to have severe impacts on the frequency and
intensity of peak electricity demand across the United States Proc. Natl Acad. Sci. 114 1886–91
[4] Meng M and Sanders K T 2019 A data-driven approach to investigate the impact of air temperature on the efficiencies of coal and
natural gas generators Appl. Energy 253 113486
[5] Sanders K T 2015 Critical review: uncharted waters? the future of the electricity-water nexus Environ. Sci. Technol. 49 51–66
[6] Dubey S, Sarvaiya J N and Seshadri B 2013 Temperature dependent photovoltaic (PV) efficiency and its effect on PV production in
the world—a review Energy Proc. 33 311–21
[7] Voisin N, Kintner-Meyer M, Skaggs R, Nguyen T, Wu D, Dirks J, Xie Y and Hejazi M 2016 Vulnerability of the US western electric
grid to hydro-climatological conditions: how bad can it get? Energy 115 1–12
[8] Burillo D, Chester M V, Pincetl S and Fournier E 2019 Electricity infrastructure vulnerabilities due to long-term growth and
extreme heat from climate change in Los Angeles County Energy Policy 128 943–53
[9] Mideksa T K and Kallbekken S 2010 The impact of climate change on the electricity market: a review Energy Policy 38 3579–85
[10] California ISO Rotating power outages (available at: www.caiso.com/Documents/Rotating-Power-Outages-Fact-Sheet.pdf)
[11] Neumann S, Sioshansi F, Vojdani A and Yee G 2006 How to get more response from demand response Electr. J. 19 24–31
[12] Zhang G, Zhong H, Tan Z, Cheng T, Xia Q and Kang C 2022 Texas electric power crisis of 2021 warns of a new blackout
mechanism CSEE J. Power Energy Syst. 81–9
[13] Andresen A X, Kurtz L C, Hondula D M, Meerow S and Gall M 2023 Understanding the social impacts of power outages in North
America: a systematic review Environ. Res. Lett. 18 053004
[14] Jones-Albertus B 2017 Confronting the duck curve: how to address over-generation of solar energy (available at: www.energy.gov/
eere/articles/confronting-duck-curve-how-address-over-generation-solar-energy)
[15] Denholm P, O’Connell M, Brinkman G and Jorgenson J 2015 Overgeneration from solar energy in California: a field guide to the
duck chart (NREL/TP-6A20-65023) Tech. Rep. p 46 (available at: www.nrel.gov/docs/fy16osti/65453.pdf)
[16] Golden R and Paulos B 2015 Curtailment of renewable energy in California and beyond Electr. J. 28 36–50
[17] Mayes S and Sanders K 2022 Quantifying the electricity, CO 2 emissions, and economic tradeoffs of precooling strategies for a
single-family home in Southern California∗Environ. Res. Infrastruct. Sustain. 2025001
[18] Mayes S, Zhang T and Sanders K T 2023 Residential precooling on a high-solar grid: impacts on CO2emissions, peak period
demand, and electricity costs across California Environ. Res. Energy 1
[19] California Public Utilities Commissoin 2021 CPUC proposals ensure electricity reliability during extreme weather for summers
2022 and 2023 (available at: https://docs.cpuc.ca.gov/PublishedDocs/Published/G000/M419/K226/419226930.PDF)
[20] California ISO Consumer conservation helps avert outages for second straight day 2020 (available at: www.caiso.com/Documents/
Consumer-Conservation-Helps-Avert-Outages-Second-Straight-Day.pdf)
[21] Bratihwait S D, Hansen D G and Hilbrink M 2013 Impact evaluation of California’s Flex Alert demand response program (available
at: www.calmac.org/publications/2013_Flex_Alert_-_Impact_Eval_-_Final_20140228.pdf)
[22] California Public Utilities Commission 2021 Order Instituting Rulemaking to Establish Policies, Processes, and Rules to Ensure
Reliable Electric Service in California in the Event of an Extreme Weather Event in 2021
[23] Yin R, Kara E C, Li Y, DeForest N, Wang K, Yong T and Stadler M 2016 Quantifying flexibility of commercial and residential loads
for demand response using setpoint changes Appl. Energy 177 149–64
[24] Chen X, Wang J, Xie J, Xu S, Yu K and Gan L 2018 Demand response potential evaluation for residential air conditioning loads IET
Gener. Transm. Distrib. 12 4260–8
[25] Dyson M E H, Borgeson S D, Tabone M D and Callaway D S 2014 Using smart meter data to estimate demand response potential,
with application to solar energy integration Energy Policy 73 607–19
[26] Ali M, Safdarian A and Lehtonen M 2015 Demand response potential of residential HVAC loads considering users preferences IEEE
PES Innovative Smart Grid Technologies Conf. Europe vol 2015
[27] Lutzenhiser L 1993 Social and behavioral aspects of energy use Annu. Rev. Energy Environ. 18 247–89
[28] Jacobsen G D and Stewart J I 2022 How do consumers respond to price complexity? Experimental evidence from the power sector
J. Environ. Econ. Manage. 116 102716
[29] Parrish B, Gross R and Heptonstall P 2019 On demand: can demand response live up to expectations in managing electricity
systems? Energy Res. Soc. Sci. 51 107–18
[30] Herter K and Wayland S 2010 Residential response to critical-peak pricing of electricity: california evidence Energy 35 1561–7
[31] Parrish B, Heptonstall P, Gross R and Sovacool B K 2020 A systematic review of motivations, enablers and barriers for consumer
engagement with residential demand response Energy Policy 138 111221
[32] Crawley J, Johnson C, Calver P and Fell M 2021 Demand response beyond the numbers: a critical reappraisal of flexibility in two
United Kingdom field trials Energy Res. Soc. Sci. 75 102032
[33] Potter J M, George S S and Jimenez L R 2014 SmartPricing options final evaluation (available at: https://efis.psc.mo.gov/mpsc/
commoncomponents/viewdocument.asp?DocId=936276747)
[34] Faruqui A and Sergici S 2010 Household response to dynamic pricing of electricity: a survey of 15 experiments J. Regul. Econ.
38 193–225
[35] Idaho Power 2006 Analysis of the residential time-of-day and energy watch pilot programs: Final report (available at: https://puc.
idaho.gov/fileroom/PublicFiles/elec/IPC/IPCE0502/company/20060329PILOT%20PROGRAMS%20FINAL%20REPORT.PDF)
[36] IBM 2007 Ontario smart price pilot final report (available at: www.smartgrid.gov/files/documents/
Ontario_Smart_Price_Pilot_Final_Report_200706.pdf)
[37] Hammerstrom D J 2007 Pacific Northwest GridWiseTM testbed demonstration projects (available at: www.pnnl.gov/main/
publications/external/technical_reports/PNNL-17167.pdf)
[38] RLW Analytics 2004 Residential time-of-use (RTOU) pilot study load research analysis report (available at: www.oeb.ca/
documents/cases/RP-2004-0203/2005-07-submissions/cdm_trccomments_toronto_supplementary2.pdf)
[39] Herter K, McAuliffe P and Rosenfeld A 2007 An exploratory analysis of California residential customer response to critical peak
pricing of electricity Energy 32 25–34
11

Environ. Res.: Energy 1(2024) 015002 M Peplinski and K T Sanders
[40] Kessels K, Kraan C, Karg L, Maggiore S, Valkering P and Laes E 2016 Fostering residential demand response through dynamic
pricing schemes: a behavioural review of smart grid pilots in Europe Sustain 8
[41] AGL Energy 2016 AGL trials impacts of emerging technologies on the grid and energy bills (available at: www.agl.com.au/about-
agl/media-centre/asx-and-media-releases/2016/march/agl-trials-impacts-of-emerging-technologies-on-the-grid-and-energy-bills)
[42] Meier A 2009 How one city cut its electricity use over 30% in six weeks European Council for an Energy-Efficient Economy Summer
Study pp 1687–91 (available at: www.eceee.org/static/media/uploads/site-2/library/conference_proceedings/
eceee_Summer_Studies/2009/Panel_8/8.020/paper.pdf)
[43] Bender S L, Moezzi M, Gossard M H and Lutzenhiser L 2002 Using mass media to influence energy consumption behavior:
california’s 2001 flex your power campaign as a case study Proc. 2002 ACEEE Summer Study vol 8 (available at: www.eceee.org/
static/media/uploads/site-2/library/conference_proceedings/ACEEE_buildings/2002/Panel_8/p8_2/paper.pd)
[44] Bartusch C and Alvehag K 2014 Further exploring the potential of residential demand response programs in electricity distribution
Appl. Energy 125 39–59
[45] California ISO Grid emergencies history report (available at: www.caiso.com/Documents/Grid-Emergencies-History-Report-1998-
Present.pdf)
[46] California Irrigation Management Information Sytem (CIMIS) CIMIS station reports (available at: https://cimis.water.ca.gov/
Stations.aspx)
[47] Environmental Protection Agency (EPA) Air Data: Air Quality Data Collected at Outdoor Monitors Across the USc (available at:
https://www.epa.gov/outdoor-air-quality-data)
[48] National Oceanic and Atmospheric Administration (NOAA) National centers for environmental information (NCEI), NOAA
NCEI local climatological data (LCD) (available at: www.ncei.noaa.gov/cdo-web/datatools/lcd)
[49] California Energy Commission Climate zone tool, maps, and information supporting the California Energy Code (available at:
https://www.energy.ca.gov/programs-and-topics/programs/building-energy-efficiency-standards/climate-zone-tool-maps-and)
[50] Office of Environmental Health Hazard Assessment 2018 C.E.P.A. CalEnviroScreen 3.0 (available at: https://oehha.ca.gov/
calenviroscreen/report/calenviroscreen–30)
[51] United States Census Bureau 2019 2021 American housing survey (AHS) (available at: www.census.gov/programs-surveys/ahs.
html)
[52] Southern California Edison Demand response programs for business (available at: www.sce.com/business/demand-response)
[53] Cong S, Nock D, Qiu Y L and Xing B 2022 Unveiling hidden energy poverty using the energy equity gap Nat. Commun. 13 2456
[54] Powells G and Fell M J 2019 Flexibility capital and flexibility justice in smart energy systems Energy Res. Soc. Sci. 54 56–59
[55] California Public Utilties Commission 2022 Advanced strategies for demand flexibility management and customer DER
compensation (available at: www.cpuc.ca.gov/-/media/cpuc-website/divisions/energy-division/documents/demand-response/
demand-response-workshops/advanced-der—demand-flexibility-management/ed-white-paper—advanced-strategies-for-
demand-flexibility-management.pdf)
[56] California Public Utilities Commissoin Power saver rewards (PSR) fact sheet (available at: www.cpuc.ca.gov/-/media/cpuc-website/
divisions/energy-division/documents/summer-2021-reliability/emergency-load-reduction-program/psr-fact-sheet.pdf)
[57] Libertson F 2022 (No) room for time-shifting energy use: reviewing and reconceptualizing flexibility capital Energy Res. Soc. Sci.
94 102886
[58] Silva S, Soares I and Pinho C 2017 Electricity demand response to price changes: the Portuguese case taking into account income
differences Energy Econ. 65 335–42
[59] Drehobl A and Ross L 2016 Lifting the high energy burden in America’s largest cities: how energy efficiency can improve
low-income and underserved communities (available at: www.ilsag.info/wp-content/uploads/SAG_files/Meeting_Materials/2017/
January_31_2017/Lifting_High_Energy_Burden_Energy_Efficiency_for_All_ACEEE_April_2016.pdf)
[60] Kwon M, Cong S, Nock D, Huang L, Qiu Y (Lucy) and Xing B 2023 Forgone summertime comfort as a function of avoided
electricity use Energy Policy 183 113813
[61] Gyamfi S, Krumdieck S and Urmee T 2013 Residential peak electricity demand response—highlights of some behavioural issues
Renew. Sust. Energy Rev. 25 71–77
[62] Trabish H K 2022 Real-time pricing, new rates and enabling technologies target demand flexibility to ease California outages
UtilityDive (available at: www.utilitydive.com/news/real-time-pricing-new-rates-and-enabling-technologies-target-demand-flexib/
631002/)
[63] Paterakis N G, ErdinçO and Catal˜
ao J P S 2017 An overview of demand response: key-elements and international experience
Renew. Sust. Energy Rev. 69 871–91
12