Principal Investigator
At a Glance
Intermittency—variability in output—bedevils wind and solar energy. Presented here is a fresh approach to intermittency that focuses on the statistics of lulls (periods of low output), especially the longest lulls. Long lulls are extreme events and should be at the center of attention in grid management. The longest lulls are unlikely to elicit the same strategies as shorter lulls. In particular, batteries (and other storage strategies that have costs roughly proportional to the energy they store) will be ill-suited to compensate for the rare, longest lulls.
Research Highlight
From 2004 to 2017, installed solar and wind power grew spectacularly on a global scale: from 3 to 300 GW for solar and from 40 to 500 GW for wind. It is far from clear that solar and wind will continue to surge, but it is time to take seriously the possibility that the world will swap the current fossil energy system for one where solar and wind dominate.
Relative to the current fossil energy system, a wind-solar energy system is more different than two other contending low-carbon systems, one dominated by fossil fuels with CO2 capture and storage (CCS) and the other by nuclear (fission and fusion) power. The intermittency and unpredictability of both solar and wind power creates fundamentally new challenges associated with long periods of low power output, or lulls. Long lulls should be conceptualized as a class of extreme events, needing dedicated attention and deserving names like those of hurricanes.
This approach explored a study of a single database: total hourly windpower production in 2016 in the region governed by the Electricity Reliability Council of Texas (ERCOT), a region approximately coincident with the state of Texas. This is a work in progress, conducted jointly by CMI PI Robert Socolow and Pedro Haro. Haro was a Visiting Fellow at Princeton University in the winter of 2017-18 and is now a professor at the University of Seville.
The question that has generated Socolow and Haro’s inquiry is: How long do periods of low power last? To answer that question, their general method is to:
- Start with a time sequence (in our case, hourly data for a year).
- Choose a threshold (in our case, less than half the annual average power).
- Measure the duration of low power (the length of the lull) from the hour the power first drops below the threshold to the hour the power first returns above the threshold.
- Create a histogram that bins these lulls by duration.
- Pay special attention to the longest lulls, treating them as distinct events.
Figure 3.2.1 displays 2016 hourly data for the total electricity production from the wind farms in the ERCOT service area. During 2016, the installed wind capacity for these wind farms climbed from 16 GW to 17.5 GW and averaged 17.1 GW, while the average windpower output was 6.0 GW. Taking into account all sources of electricity, the average ERCOT demand in 2016 was 40.0 GW, so average windpower production was 15% of average demand.
Socolow’s and Haro’s choice for the threshold defining “low” output (horizontal line in Figure 3.2.1) is half of the annual average value, or 3.0 GW.
Figure 3.2.2 arranges all lulls by their duration, and plots the total time of lulls for each duration. The four longest lulls are labeled A, B, C, and D in the histogram, and these events are also identified in Figure 3.2.1. Event A lasts roughly four days, and events B, C, and D last roughly two days—a total of 10 days for the four events.
Figure 3.2.3 reveals that altogether in 2016 there were 219 lulls accounting for 2059 hours (nearly one quarter of a year), and the average duration of a lull was nine hours. However, averages over lulls carry very little information. More importantly, 75% of the lulls lasted for less than 12 hours and 95% for less than a day. There were only 12 long lulls, defined as lasting more than a day, yet these accounted for 481 hours (20 days).
| Number of Events | Cumulative Hours | % of Total Hours | ||
| Lull Length (hrs) | <12 | 164 | 862 | 42 |
| 13-24 | 43 | 716 | 35 | |
| 25-36 | 8 | 247 | 12 | |
| >36 | 4 | 234 | 11 | |
| Total | 219 | 2059 | 100 | |
Figure 3.2.3. A table of summary statistics for these lulls.
Other data sets. We examined the same ERCOT data using a lower threshold, 25% of annual average power, or 1.5 GW. The longest lull with this tighter definition lasted less than half as long, 20 hours.
The researchers also examined one other database, total hourly windpower production in mainland Spain in 2016, with a threshold of 50% of average power, and found shorter lulls than for ERCOT: the maximum lull was 17 hours, and the second longest was 11 hours.
For sure, this methodology could also reveal important features of “solar lulls,” stretches of cloudy days. If applied to solar-wind lulls (the periods of low output when both solar and wind power are contributing to a single power system), it would show the degree to which solar and wind in combination could shorten and make less deep the lulls that either would produce independently.
Implications for the management of intermittency. Lulls need to be partitioned for optimal management. The longest lulls are unlikely to elicit the same strategies as the shorter ones. In particular, batteries, which have costs roughly proportional to the energy they store, will be ill-suited to compensate for the longest lulls. In this example, credibly, lulls under 12 hours, and perhaps in the future under 24 hours, would be matched to batteries, while the eight events lasting between 25 and 36 hours, the three two-day events, and the single four-day event would require other solutions.
Ongoing analysis. Prolonged low wind is the obvious explanation for long lulls in windpower output. However, it is proving not at all straightforward to identify events directly from wind speed data, even when data are combined for several locations in ERCOT territory. Two explanations that Socolow and Haro have been able to exclude are: 1) edicts from ERCOT leading to curtailments, and 2) extensive storms creating winds exceeding turbine cut-out speeds.
Future work. It will be important to understand the extent to which long lulls can be predicted in advance; having warnings of long lulls would improve preparedness (use of reserve capacity, expansion of imports, and shutdown of non-essential activity). Investigations of spatial correlations specific to long lulls will improve assessments of the value of long-distance transmission capacity and clarify the trade-offs between back-up power (e.g., provided by natural gas or diesel generators), storage, and demand-side management. Long-lull analysis should also provide insight into likely behavioral responses to these stresses.
Acknowledgments
Pedro Haro is a full co-author, with Robert Socolow, not only producing clever data displays but also contributing key hypotheses. Both authors wish to emphasize that this work is at an early stage.
The inspiration for this work was a field trip in May 2017 to bp’s Sherbino 2 wind farm in west Texas, hosted by Jason McDonald of bp and involving eight Princeton participants (see Figure 3.2.4). That particular day the team experienced dramatically variable wind: the wind, having died down after midnight, was recovering briskly as we arrived in the early morning. McDonald has continued to be helpful and resourceful. Haro and Socolow are also benefitting from discussions with Paul Wattles, Aaron Townsend, Pengwei Du, and Kevin Hanson at ERCOT.
References
Haro and Socolow found two earlier papers that investigated the frequency of lulls vs. their duration, for lulls in either wind velocity or windpower:
Kaldellis, J.K., 2002. Optimum autonomous wind-power system sizing for remote consumers, using long-term wind speed data. Applied Energy. 71: 215–233. doi: 10.1016/S0306-2619(02)00005-3.
Poder, V., T. Peets, K. Toom, and A. Annuk, 2010. The estimation of wind lull and consumption factor influence on autonomous wind energy system. Agronomy Research. 8: 226-235, doi: 10.3176/oil.2009.3S.10.