IEEE Power & Energy Magazine - November/December 2017 - 53

DER Impacts CAISO's
Ability to Forecast Load
Load forecasting is essential to market optimization and
system reliability. Traditionally, load forecasting was driven
by end-user consumption patterns, such as people waking
up, leaving home for work, and returning home after work.
With DPVs, reliance on the grid is driven both by end-user
consumption patterns and changing DPV output. Now, PV
cloud cover (irradiance) forecasting has become a critical
element in the load forecast, adding to other meteorologi-
cal components that were already inherent in demand, such
as temperature and humidity. Load forecasting models are
trained using actual historical data. Granularity of data is
important so that the model can simulate individual effects
to train the load forecast. Because California does not meter
the actual generation from BTM PVs, it is challenging to
effectively train the load forecast model. Currently, CAISO
is using actual historical irradiance values for estimating
DPV output to train the models, but inaccuracies in this esti-
mate add to the load forecast error.
At the end of 2016, CAISO had 4,903 MW of BTM PVs.
This is expected to double by 2021. For a system that peaks
at about 50 GW, this is significant penetration.
Figure 4 shows seasonal graphs of one of the load fore-
casting areas in California. The graphs depict how the load
shape has transitioned from 2010 to 2016, with 2010 being
the lightest shade and 2016 being the darkest.
november/december 2017

1,200
1,000
800
600
400

12 a.m

9 p.m.

6 p.m.

3 p.m.

12 p.m

9 a.m.

12 a.m.

6 a.m.

200
3 a.m.

System Load (MW)

imbalances exceeding downward reserve and minimum
output levels (with all utility-scale renewables fully cur-
tailed). This will lead to grid instability if inverter capa-
bilities such as frequency and voltage ride-through, droop
functions, and communication, command, and control pro-
tocols are unavailable.
Figure 3 shows the evolution of HECO's system load dur-
ing the period 2007-2017 due to increasing levels of BTM
DPVs. DPVs shift bad changes the system's ramping needs.
Pre- and post-2012 (when DPVs grew exponentially), the
largest ramp shifted from the morning to the afternoon. On
many days, this ramp has become steeper and more com-
pressed in time than before. Additionally, system operators
must manage more ramp events during the day; in 2017,
for example, the Oahu system saw the daily minimum load
occur during the day, instead of at 3:00 a.m., which is when
it traditionally had occurred. Finally, it impacts the utility's
ability to accurately forecast the load, because DPVs is able
to serve a significant fraction of the load during certain
hours and on certain days.
Hawaii continues to promote DERs that have charac-
teristics of conventional generators. After requirements to
expand voltage and frequency ride-through settings in 2014,
HECO is currently evaluating frequency droop capabilities
for inverters, with the ultimate goal of enacting standards
and infrastructure that enable dispatchable DERs for system
reliability needs.

Time of Day
2018 Gross Load
2018 Net Load (No DPV)
2018 Net Load (Current DPV)
2018 Net Load (Preapproved DPV)
2018 Net Load (Two Times Current DPV)
2018 Pmin + Reserve

figure 2. HECO's "duck curve," showing increased impacts of DPV on system load for the worst day of the year.

1,200

HECO
Historical Comparison of System Load (MW)
Fourth Week of January-Week 4

1,100
1,000
900
800
700
600
500
Mon.

Tue.

Wed.

Thu.

Week Beginning
22 January 2007
19 January 2009
24 January 2011
21 January 2013
19 January 2015
23 January 2017

Fri.

Sat.

Sun.

21 January 2008
25 January 2010
23 January 2012
20 January 2014
18 January 2016

figure 3. The evolution of the HECO system load, showing
the impact of BTM DPV during January from 2007 to 2017.

Load forecasting models typically perform best with sig-
nificant amounts of historical data. However, rapid changes
in the load shapes [see Figure 4(a) and (c) for fall and spring,
respectively] make it difficult to use even the previous year's
data for training.
ieee power & energy magazine

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2017