Using Multiple Metrics
Most utilities will pick a single resource adequacy metric as the target for their LOLE studies when the goal is to identify a single planning parameter like PRM. However, evaluating a system via multiple metrics provides a fuller picture of its resource adequacy. Hawaiian Electric, for example, conducted a resource adequacy study on its proposed plan that quantified all four of these metrics across multiple scenarios: 1) events/year, 2) days/year, 3) hours/year, and 4) unserved energy . EPRI’s ongoing RA effort is examining this aspect and looking at how different metrics could be used to provide greater insights.
In a system with an increasingly diverse supply and demand, the previously held assumption that a close positive correlation between LOLE and EUE may no longer hold valid as the net load profile becomes increasingly variable. A more explicit effort to examine frequency, duration and magnitude metrics concurrently is warranted. As seen in , Figure 5-2 illustrates how different reliability metrics may be complementary, describing different dimensions of system adequacy for a specific event. Each blue block represents a quantity of demand that is shed due to inadequacy. The dashed PRM line indicates a fictious planning reserve margin above the peak demand forecast for that system, expressed in MW. PRM has been added to the Figure to illustrate how it remains static, giving information on the capacity of a system to meet peak demand and not its expected performance.
Figure 5-2 a and c show two events where load is lost for three hours. LOLEv and LOLH indices show the same values for these two very different events, EUE however uncovers that even if both of these are single events occurring for three hours, the supply shortage is three times greater in event ‘a’ than in event ‘c’.
Similarly, Figure 5-2 b and d show two distinct instances where the expected unserved energy is 6 MWh for both, with 3 hours of supply adequacy issues in both cases. Considering LOLEv shows that, while loss of load occurs in one single event in case ‘b’, there are three separate loss of load events in case ‘d’.
While PRM and LOLE metrics have predominated, combinations of the presented indices may be used going forward into power systems with less dispatchable and energy-limited generation, and, generally, higher flexibility needs. LOLE and PRM metrics may have been adequate in traditional power systems, dominated by thermal generation, however, trends point to an increase in the variety of reliability events (i.e., not limited to peak load hours). Furthermore, while universally recognized as an RA metric, the precise calculation methodology for the LOLE metric varies between implementations and regions for the same underlying model and therefore further explanation is warranted for what exactly is being calculated in any RA study.
The 2020 California and 2021 Texas events have also shown that, although probabilistic metrics have been calculated in the past based on expected (average) values, accounting for the distribution of shortfall events, as well as high impact low-frequency events, is expected to become increasingly relevant going forward. This is particularly relevant as the frequency of extreme events is expected to increase resulting from climate change, at the same time as generation is becoming increasingly weather dependent.
Finally, it is recognized that expected values portray the adequacy risk across a wide number of potential outcomes. The distribution of reliability outcomes in each weather year, outage draw, or scenario will vary across a distribution. In some cases, the variance of that distribution will be relatively small, but reliability outcomes may be more diverse in other distributions. A long-tailed distribution of adequacy outcomes in each scenario may also be plausible. While expected values may demonstrate an acceptably low risk on average, tail events may present unacceptable damage potential to society.
As a result, certain regions incorporate metrics accounting for lower probability events. For example, in addition to maintaining a LOLE below 3 hours per year, Belgian regulations mandate that the LOLE95, or the loss of load expectation at the 95th percentile of the cumulative density function of the loss of load expectation, is kept below 20 hours per year. By doing so, two metrics are used to assess both expected and tail risks with acceptability criteria for both.