“Can resource adequacy be attained without defining what is ‘enough’?” This is the astute question posed by Beth Garza, formerly Independent Market Monitor for ERCOT and now senior fellow at the R Street Institute think tank. In this blog, I would like to engage with her question.
Customary short-hand descriptions of resource adequacy focus on installed reserve margin, which is the amount by which the total power generation capacity exceeds a forecast peak consumption. I will argue that, in a high renewable world, the focus on power capacity over a short time interval at the time of a forecast peak is not a suitable short-hand, because adequacy will become more dependent on the availability of energy over an extended time. The “what” in Beth Garza’s question will increasingly need to be thought of as energy capacity rather than power capacity, and we will need to define how much energy capacity is needed to satisfy our requirements for adequacy.
To understand this change in the needed short-hand for adequacy, let’s first think about assessing resource adequacy in systems with mostly thermal generation. Typically, load is at peak levels for just a few hours in summer or winter. In thermal-dominated systems, resource adequacy is roughly tantamount to having enough thermal generation capacity available with high enough probability to meet a particular future peak load condition. Outside of these peak load hours, there is generally sufficient capacity to meet load, even considering failures and the need for annual maintenance.
There are various considerations in an assessment of adequacy in a thermal-dominated system that hinge on uncertainties and probabilistic assessments. On the demand side, probabilistic assessments arise because we must forecast future peak load conditions, including the extremity of associated weather conditions that drive both winter and summer peaks. In other words, there is an uncertain future peak load, so the definition of resource adequacy must consider how extreme the peak.
To put it another way: in order to define whether resources are adequate, we must specify the forecast load that the resources are supplying. Implicitly, there is a non-zero probability that the actual realized peak load exceeds the forecast peak load. For example, peak loads in the February 2021 event in ERCOT exceeded the ERCOT assessment of forecast peak loads for winter 2020-2021, because the extreme weather that actually occurred in February 2021 was a once-a-decade phenomenon. The forecast peak considered only more typical winter peaks.
It is not just load that has randomness. Generators, too, have random failures. The assessment of resource adequacy must therefore also consider the probabilities of failure of thermal generation. Historical statistics are typically used to estimate generator failure rates.
Putting the demand and supply together, a specification of resource adequacy must define the minimum acceptable probability for being able to supply all load. This could equivalently be described as deciding how far out to consider on the “tail” of unlikely events of peak load variation and generator failures. Given the minimum acceptable probability of being able to supply all load all the time, we can assess whether or not the resources are adequate. A typical minimum acceptable probability of supplying all the load might be 99.97% over a year.
To summarize, the question about thermal resource adequacy typically comes down to a question about the likelihood of power production capacity being available to meet peak power consumption conditions. This assessment primarily depends on a particular, relatively small, length of time during load peaks and considers the probability distribution of power capacity in relation to the probability distribution of peak load. This is appropriate for a predominantly thermal system with “peaky” demand, where failures of thermal generation are uncorrelated from generator to generator, and where the critical demand periods are particular hours sporadically occurring over a summer or winter, with the occurrence of these peaks uncorrelated with generator outages.
So how can we evaluate whether there are enough thermal generation resources to satisfy our specification of adequacy? One approach is to define the concept of “effective load carrying capacity” (ELCC) of each generation resource. For thermal resources, if their failures are not correlated with demand conditions and the failures are uncorrelated across generators, then the ELCC can be roughly evaluated as the installed capacity of the generator derated (reduced) by its failure or forced outage rate. The derated capacity can be viewed as the “expected” availability in a probabilistic sense.
Adding up the capabilities of a large number of generators, and assuming that failures across generators are uncorrelated and that failure rates do not change over time, the law of large numbers tells us that the sum of the actual available capacities of all the generators will be roughly equal to the sum of these derated capacities. To put it another way, we can roughly think of the derated capacity as representing the capacity of an equivalent perfectly reliable generator that is always available. Adequacy is tantamount to having enough equivalent perfectly reliable generation capacity.
How can we interpret this in terms of installed reserve margin? Adding up the total installed capacity in a system with adequate generation, we will find that it exceeds the peak load forecast. That is, there is a reserve margin. Historically, an installed reserve margin of around 12 to 15% above the peak load forecast provided adequate capacity in a thermal-dominated system.
A more refined calculation considers the distribution of failures more carefully to evaluate the derated capacities. Stanford Professor Frank Wolak provides some examples in his paper “Long-Term Resource Adequacy in Wholesale Electricity Markets with Significant Intermittent Renewables.” If the sum of the derated capacities is sufficient, then the assessment is that supply would be able to meet load with a probability that is at least the minimum acceptable level. If not, then significant involuntary curtailment of load would be required or new generation should be built. In a resource adequacy context, the potential for significant curtailment would point to the need to build new generation before the season of these forecast load peaks with a view to increasing the reserve margin sufficiently.
A complication with this analysis relates to generator failure rates. In fact, “common mode events,” such as extreme cold or heat, can increase the failure rates of generators, as experienced in the ERCOT February 2021 event and therefore mean that there is some correlation of thermal generation failures with weather and load. This issue is discussed at length in an EPRI report and in my blogpost on the ERCOT event. In principle, this effect can be included or approximated in the analysis.
How do renewables change this situation? Unlike thermal generators, the availabilities of renewable resources are correlated from one resource to another and also highly correlated with weather conditions and load. When it is windy at one wind farm in west Texas, it is likely to be windy at most West Texas wind farms, and when it is not windy at one wind farm, it is likely to be not windy at most wind farms. This correlation means that the law of large numbers cannot be used in the same way as for thermal generation. It invalidates the idea of “adding together” derated capacities of individual wind farms since the availabilities are not independent across farms.
To consider these correlations, one approach is to consider the net load, the demand minus total renewable production. This necessitates forecasting simultaneous renewable production and demand, including the extremity of the weather conditions. Adequacy comes down to whether the thermal generation and storage can meet the forecast net load, bearing in mind that the time of the net load peak will differ from the time of the load peak.
Modern ELCC software can evaluate these situations. However, interpretation of the results for renewables is different to the notion of capacity derating for thermal generation. For a thermal generator, we expect that its ELCC will be roughly, although not completely, independent of what other resources are built or retired and roughly independent of how other resources are operated. This is consistent with the “rule of thumb” of a 12 to 15% installed reserve margin being adequate in a thermal system.
In contrast, the ELCC for a particular wind farm calculated for, say, the case of 30GW of installed wind capacity will be significantly lower than the ELCC calculated for the case of 20GW of installed wind capacity and the ELCC can depend significantly on the other available resources such as storage and how they are operated. As renewable penetration increases, the correlation of production across renewables implies that the ELCC per MW of installed capacity will decrease. For example, in a 2019 study by Energy and Environmental Economics (E3) of deep decarbonization for California, E3 expects ELCC for solar farms to fall from about 50% of farm capacity to around 1% of farm capacity as the penetration of solar increases significantly. This means that a particular level of installed reserve margin will no longer be a suitable short-hand for adequacy in a renewable-dominated system because the reserve margin necessary to achieve an acceptable level of adequacy will be highly dependent on the assets in the system.
System operators recognize this issue and can consider it in their calculations. Again, Professor Wolak provides a detailed explanation of the process. ELCC assessments of resource adequacy could then, in principle, use derated capacities of individual thermal resources together with a derated total renewable capacity to assess whether there will be enough available capacity at the time of a forecasted future peak net load. Professor Wolak details some of the serious technical difficulties in trying to apply ELCC in high renewable contexts.
Professor Wolak’s critique of ELCC applied to renewables and the discussion in various reports, including the E3 California report and work by the Energy Systems Integration Group, point to why adequacy cannot be captured in high renewable systems by power capacity concepts such as static levels of installed reserve margins. Installed reserve margins implicitly reference a peak load or peak net load condition; however, under high renewable penetration, adequacy is increasingly determined by supply-demand balance during the extended periods of low renewable production that the Germans call a dunkelflaute. While sophisticated simulations can reflect these sorts of energy constraints, translating them into capacity terms such as an ELCC or a required installed reserve margin then obscures the underlying energy issues.
Correlation of renewable production over multiple hours or days brings into question the whole power capacity focus of reserve margin assessment. It is not simply that there might be low renewable production during the particular hour or few hours of peak load or peak net load in a summer or winter and that we might or might not have enough available capacity for that hour or few hours. The issue is more serious: there might be low renewable production for many hours, or even days, resulting in a significant mismatch between the desired energy consumption and the available energy production.
This was illustrated by the February 2021 event in ERCOT, which incidentally also involved correlated outages of generation and natural gas supply due to cold weather. While winterization will reduce the coincidence of future outages of thermal generation and gas supply under cold conditions, and more generally reduce the prevalence of other issues that can be mitigated by winterization, it will not change the correlation between lowered renewable production and increased consumption, since these are due to the “common mode” event of the extreme weather event itself.
Consider a future repeat of a similar weather event to February 2021 in ERCOT. Let’s assume winterization of the thermal generation, gas supply, water supply, and other components has been accomplished. And, suppose similar weather conditions occur in a future ERCOT with much higher penetration of renewables.
ELCC assessment that included this particular weather event would reveal whether or not the other generation and storage in the system could meet demand or whether significant curtailment was necessary. Although repetition of something like the February 2021 weather might be a low probability event, its severity would lead to significant hardship. That is, the most pressing question for future adequacy will increasingly be whether there is enough energy over multiple hours to days to cover an event similar to the February 2021 ERCOT event without significant curtailment. This energy would come from a combination of renewable production, available fuel at thermal generators, and from other storage such as batteries. From this perspective, the recent discussions in ERCOT around new mechanisms for ensuring adequate generation capacity seem beside the point: these discussions all build on an anlysis that did not consider the 2021 weather conditions.
The ERCOT market has recently added a “Firm Fuel Supply Service.” Requirements for onsite fuel storage for thermal generators is a tacit recognition that power capability is insufficient to evaluate resource adequacy when stressed conditions may extend over multiple contiguous hours or days. The issue of supplying energy over multiple contiguous hours or days will become increasingly significant at higher levels of renewables.
So what should be used to assess resource adequacy moving forward in ERCOT? I believe that we will need to move toward the sort of evaluations that have been used in hydro dominated systems such as Brazil, Chile, New Zealand, and Tasmania in Australia. Such studies do not focus on a particular hour or a few hours. Rather, they consider the various scenarios of renewable availability and storage over extended times. That is, they consider the probability distribution of energy capacity in fuels and storage to assess whether or not the load energy can be met at the required minimum acceptable probability.
Seasonal water inflow and water storage in Brazil, Chile, New Zealand, and Tasmania dictate that the timescales for assessment in these countries are on the order of months to years. ERCOT has very little hydro and no pumped storage hydro. Therefore, the timescales relevant to ERCOT are shorter due to the need to compensate for daily or weekly renewable fluctuations instead of seasonal water inflow. Nevertheless, ERCOT will need to consider scenarios of load and of generation that unfold over many hours to days.
This type of analysis is familiar to operators of hydro-dominated systems but full assessment may require further evolution of assessment tools for ERCOT, or at the very least require consideration of extreme weather events within existing ELCC tools. We need to make these informed assessments before we have another curtailment event like February 2021. The reforms of the ERCOT market that are being discussed currently should be thoroughly analyzed to determine if they result in appropriate levels of resource adequacy considering the emerging energy dimension of resource adequacy.
LOLE or loss of load expectation is the annual sum of daily probabilities of not serving the load during each day. With renewables as part of the supply the peak loss of load probability could occur at other hours other than the peak demand hour.
The NERC objective is to keep the LOLE < 0.1 days/year.
If we wish to have renewables supplying larger amounts of energy we definitely have a need to know if those renewables are going to produce enough energy. I don't know of any reliability measure that indicates how much risk there is in the ability to generate enough energy. Possibly a new definition is needed.
Even if we do invent a new definition to calculate the energy supply reliability, it still doesn't solve the problem of predicting its frequency of occurrence. With generator outages we have a reliable random outage prediction tool. But we don't have a reliable weather predictor for future years.
If we have Uri events every ten years it suggests we need to start planning for them.
Ross I’m listening to an ERCOT meeting 9:30 AM Mar 15 on their reliability modeling
using a Monte Carlo program. Also mentioned in the meeting is that NERC is working on a new energy adequacy metric. I’m not in the loop with NERC so I don’t know the details. NERC should have some kind of energy metric soon. Maybe we can find out what they have in mind. Let me search. Yes there are search hits on this https://duckduckgo.com/?q=nerc+energy+adequacy&t=h_&ia=web