IEEE Power & Energy Magazine - May/June 2018 - 24
Principal Component 3
ipal C -4 -6
-5 0 t 1
figure 7. A scatter plot of three principal components of
20 voltage variables recorded in the Irish networks for case 2.
Lead Time (h)
Pred. Intervals (Cov. Rates from 10 to 90%)
uncertainty has always been around in power system
operation and planning. for example, operational decision
and control problem uncertainties originate from contingencies (generation units and lines), incomplete or erroneous overviews of the system state, and projections of
future demand. today, however, with the rapid deployment
of renewable energy generation capacities throughout the
world, new uncertainties are appearing that directly relate
to how much power may be generated in upcoming minutes, hours, days, and beyond. similarly, on the electricity
consumption side, uncertainties are growing due to changes
in consumption patterns (such as electric vehicles and more
proactive consumers) but also to behind-the-meter power
generation. combined with an all-time high availability
of relevant data, this has supported the increased focus on
developing new approaches to analytics and forecasting for
power system operations and control.
While traditional point (or single-valued) forecasts can
provide the expected values for the variable of interest,
probabilistic forecasts, which have been around for more
than a decade, can further quantify the future uncertainties via quantiles, intervals, or probability distributions. it
is challenging to visualize such uncertainties so that the
probabilistic forecasts can be effectively communicated
to and ultimately accepted by the business consumers of
these forecasts. in this section, we will introduce and discuss alternative approaches to visualizing probabilistic
wind power forecasts.
River-of-Blood Fan Chart
figure 8. Probabilistic forecasts represented as a riverof-blood fan chart, with a decreasing shade intensity for
higher nominal coverage rate of the prediction intervals,
for the whole wind power generation of western Denmark,
with an hourly resolution up to nearly two days ahead.
test. as illustrated in the scatter plot of the three principal
components in figure 7, the original steady state is represented by the yellow dots; as the event progresses, it goes
from the black dots to the red ones and the blue ones and
finally settled to a new steady state represented by the green
dots. the spiral trace indicates the oscillatory behavior
during this test. the graphical visualization in figure 7
provides a faster and easier way to interpret information,
which helps reduce the decision-making time.
Visualizing Probabilistic Forecasts
While visualizing the data at the beginning of data analysis
is well known to be a required step, an equally important
step is visualizing the results from sophisticated models.
Here we will present another case study, focusing on the
visualization of forecasting results. We will use wind power
forecasts as an example, although the methodology can be
generally applied to other energy forecasts, such as solar
power and load forecasts.
ieee power & energy magazine
a prominent example of communicating probabilistic forecast information is through a "river-of-blood" fan chart, as
depicted in figure 8. an earlier version has been used, since
2005, in a significant number of technical presentations and
broad-audience articles to introduce and illustrate the concept
of probabilistic wind-power forecasting. this plot illustrates
hourly power generation from wind power (in this case, for the
whole wind power generation of western Denmark), with an
hourly resolution up to nearly two days ahead. this visualization proposal is inspired by the Bank of england's probabilistic forecasts for inflation, which have published on a quarterly
basis from 1996, as a pragmatic and intuitive approach to convey uncertainty information.
this so-called river-of-blood fan chart associates the
traditional single-valued forecasts, relaying the mean of
potential renewable power generation in the near future (formally, the conditional expectation) with a number of prediction intervals. these prediction intervals have an increasing
nominal coverage rate, hence intuitively getting wider for
lighter colors. for a given lead time, a prediction interval
gives a range within which power generation may lie, given
a certain a priori probability, i.e., its nominal coverage rate.
those prediction intervals are centered in probability on the
median. the visualization appeals to both a broad audience
and expert practitioners. the former may be content with a