IEEE Power & Energy Magazine - November/December 2019 - 56

One of the biggest hurdles for implementing probabilistic forecasting
in industry is the lack of case studies, as well as reference materials
in the form of documentation, standards, and guidelines.
period by using the Poisson distribution. Once the probabilities for all groups expected to run the next day are
calculated, a Monte Carlo simulation is performed that
provides the combined pdf [Figure 11(c)]. The sum of the
three previous stochastic variables is another stochastic
variable that indicates the reserve necessary to cover the
demand [Figure 11(d)].
Figure 12 displays the amount of reserve required for each
horizon with its margin of confidence. In this way, periodic
runs of the probabilistic method determine a secure reserve
level. REE also calculates reserve based on a deterministic
method, which uses the confidence intervals of the wind
forecast (P50 and P85) calculated from meteorological
ensembles of type 4 (see Figure 1). In this way, the reserve
is computed as the sum of the maximum loss of generation
due to simple failure (1,000 MW), demand deviation (2%),
wind power deviation (difference between P50 and P85),
and additional reserve. The probabilistic method of sizing
the reserves allows the system to carry less reserve without
affecting the security of the supply, which is the highest priority for REE.

Experimental and Future Uses of
Probabilistic Forecasts
Probabilistic forecasts can be used to enhance power system stability and forecast future system conditions. As
6,000
5,000

80%
96%

85%
97%

90%
98%

95%
99%

VER penetration increases, reserve capacities decrease
while ramping increases. Look-ahead system assessment
tools will become more important to maintaining an economic dispatch structure and unit commitment.
For instance, probabilistic forecasts are being designed for
application in future California Independent System Operator (CAISO) flexible ramp products (both day ahead and real
time). Currently CAISO has deployed probabilistic method
type 1 (see Figure 1) for a real-time flexible ramp product.
This method leans heavily on a statistical histogram methodology that examines historic forecast movement between
two periods of time and sets a requirement based on the
uncertainty of net load. In the future, CAISO plans to expand
this to include forecast information, either as regressors in
the quantile regression technique (type 2) or using the
ensemble approach (type 3 or 4). CAISO also plans to include
a new commitment called day-ahead reliability and deliverability assessment (replacing Residual Unit Commitment) that
will allow multiple scenario optimization beyond the meteorological information.
The Australian Energy Market Operator developed a
Bayesian belief network to assess historic forecast performance and then project forward, dynamically, using forecast
conditions. This augments the reserve assessments based on
the usual asset contingencies. This approach considers that
weather forecast error (wind, solar, and temperature) and
contingent weather events show an
increasing impact on the network
and at times present the largest
contingency other than historic
rare largest-asset contingencies.

Summary

(MW)

4,000
3,000
2,000

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

1,000

Horizons

figure 12. A graph from the REE system of probabilistic reserves at different time
horizons and different confidence intervals.
56

ieee power & energy magazine

Because the power system is sensitive to weather information and
the atmosphere is inherently variable, it is prudent to quantify the
uncertainty in weather forecasts
and leverage that probabilistic
information to the best advantage
of system operators. Here, we have
listed current methods to provide probabilistic forecasts and
discussed how to judge whether
the probabilistic forecast is sharp,
resolves different distributions,
and is statistically reliable. We
november/december 2019

IEEE Power & Energy Magazine - November/December 2019

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2019