Obtain the data you need to make the most informed decisions by accessing our extensive portfolio of information, analytics, and expertise. Sign in to the product or service center of your choice.
When evaluating potential project sites early in the development
cycle, energy storage developers are faced with a difficult
question: how to forecast or characterize the Locational Marginal
Price (LMP) that will determine the energy arbitrage margins of the
asset. Arbitrage margins are dependent on the intraday LMP spread
and require a forecast methodology that captures both the zonal
price spread and the volatility of the LMP basis, the difference
between the zonal price and the price at the interconnection
bus.
The cost and time required to calibrate a long-term nodal
dispatch model across hundreds or thousands of nodes makes it
impractical for early stage prospecting activities, so developers
typically default to using hourly zone or hub price forecasts
supplemented with crude heuristic assessments of LMP basis
constructed using historical data. Commonly used heuristics include
straight averages, peak/off-peak averages, peak/off-peak by season
averages, or month-hour averages. Constructing these heuristics
from historical data is still a manual process, an analyst needs to
review the available historical LMP basis data and select only the
data that they believe represents the current state of the grid
surrounding the node. Large generators coming online nearby or
transmission upgrades can dramatically shift the LMP basis at a
node. The principal limitation of even the best heuristics is that
they understate the volatility of LMP basis and necessarily
understate the value captured by energy storage arbitrage
scheduling.
Applying the best performing heuristic, month-hour, against a
historical period for a high spread node in ERCOT-South allows us
to understand the magnitude of the impact that these methods have
on assessing arbitrage margins. Margins calculated using a
heuristic are understated by 26%, only slightly better than those
calculated using zonal price alone (32%). FastLMP, IHS Markit's
package of statistical and machine learning models, is able to get
within 5% of the actual arbitrage margins for this bus.
Extending the analysis to include a node just a few miles away
with a much lower spread, and examining the forecast rather than
the historical period, reveals the wide gap between the margins
forecasted by FastLMP and those calculated using a heuristic or
zonal price series. The difference between low and high spread
nodes also illustrates the importance of site
selection—identifying the right location for a storage project
can increase arbitrage margins substantially. FastLMP expedites and
facilitates this identification process.
FastLMP is able to replicate the volatility of LMP basis at a
node better than any heuristic or single model, while
simultaneously reducing absolute error, by selecting from a bundle
of statistical models that identify relationships between LMP basis
and fundamental market variables like wind, load, delivered natural
gas prices, and price differences among neighboring zones within a
power market. FastLMP also has a stochastic layer that strips
outage events that can distort heuristics and reinjects them into
the forecast considering sign, duration, magnitude, and
chronological variables. FastLMP relies on IHS Markit forecasts of
these fundamental variables to extend these trends into the
forecast.
Duncan Anderson, Senior Research Analyst at IHS Markit,
focuses on the development and operation of market simulation, big
data, and advanced analytical models for the North American Power
Analytics team.
Barclay Gibbs, Senior Director of Power and Renewables
at IHS Markit, specializes in power market analysis, due diligence,
and regulatory advisory for the North American Power and Renewables
team.
Sam Huntington, Associate Director with the Gas, Power,
and Energy Futures team at IHS Markit, focuses on energy storage
and power market fundamentals.
Posted on 15 September 2020
Learn more about FastLMP in our in-demand
webinar:
