Load Models for Loss-Of-Load Calculations

Draft revised 01/19/10

Load Model Issues for Loss-of-Load Calculations

Introduction

The NERC Planning Committee’s G&T Reliability Planning Models Task Force (GTRPMTF) is to “develop a common composite generation and transmission reliability modeling methodology for the purpose of assessing system resource adequacy, which considers the ability of load to receive power supplied by aggregate resources.”[1]

Preliminary Recommendations

The task force has not finalized its recommendations on the details of the methodology. However, these are some preliminary recommendations:

1. A probabilistic analysis that will cover all hours of the year; e.g., an 8,760-hour model. The analysis should be able to include the extra 24 hours for leap year. The calculation of probabilistic indices (such as loss of load hours - LOLH) across all hours will:
2. Include the random outages of supply side resources,
3. Allow the modeling to reflect transmission limitations, thereby preventing the inclusion of unavailable/undeliverable capacity as supplying demand.
4. Accommodate modeling of energy-limited/variable resources ;
5. Be able to address load shifts to near-peak hours as a result of demand response energy payback.
6. Develop reliability models for each year of the next ten years covered by the LTRA reporting period.
7. A common modeling methodology needs to be used within an Interconnection. The reason is that calculation of probabilistic indices for a “system” will include the explicit consideration of assistance from neighboring systems. Therefore, all systems need a common modeling methodology within an Interconnection to achieve that goal.[2] [The biggest issue is the amount of capacity assistance to be assumed to be available and deliverable from external sources.] It is envisioned that a Monte Carlo model would be used. There are two ways generally used to include transmission system characteristics in the Monte Carlo simulation. They differ in their representation of the transmission system.

a.  A transportation model represents zones of generation and load, which for modeling purposes, have no internal transmission constraints. All generation and load in a zone can be modeled as connected to one electrical bus. Zones are connected by a single line representing the equivalent transfer limit, with constraints modeled with various levels of detail. Power flow on these representative lines does not follow electrical laws of physics but instead are modeled as “transportation” e.g., roads, water, etc.

b.  A dc power flow model represents the transmission network at the bus level – loads are defined at individual busses and generators are connected to individual buses. Power flows across the transmission system according to the electrical laws of physics. The thermal impact of transmission outages must be considered explicitly in the reliability calculations because the flows over any element cannot exceed N-1 even though an all-lines-in-service (N-0) calculation would remain within satisfactory thermal limits during pre-contingency conditions. Constraints on the transmission network may also be enforced to limit the electrical impact of voltage and stability concerns resulting from contingencies. However, because the power flow uses a “dc” model and not “ac” model, actual power flow restrictions due to limited var capacity and therefore voltage based constraints are not modeled.

Load Model Issues

Two load modeling issues have been formulated by the task force.

1. Load models need to be developed in a coordinated manner within each Interconnection so that the results of the individual system load models “make sense” when they are aggregated at three different levels:
2. The Interconnection level. The aggregated subregional load shapes should be sensible at the Interconnection level.
3. The subregion level. For example, WECC has four U.S. subregions, each of which includes many systems. Each subregion’s load shape should maintain a sensible chronological relationship with those of its neighboring subregions.
4. The transmission constrained level. For transportation models, there needs to be a process to allocate the hourly loads at a system level to transmission constrained areas that may encompass one, two or more reporting entities as well as portions of a reporting entity. This could be alleviated by a granular, bus level, forecast that could then be “rolled-up” into a transmission constrained area. However, the bus-level forecasts would need to sensible when summed at the subregion and Interconnection level.

It has been observed that the system loads are driven by common weather patterns that impact multiple electric power systems across large geographic regions simultaneously. Without coordination of the load models, the peak days selected by adjacent systems may not reflect the historic diversity/correlation experienced between the systems when all systems are numerically combined. Hourly aggregated load models need to be able to assign forecasted hourly loads to zones (for transportation models) or to busses (for dc power flow models).

1. The second issue is the representation of load forecast uncertainty. While the determination of what uncertainties to model has not been confirmed by the GTRPMTF, its present thinking is:
2. Weather uncertainty is considered over the entire calculation period, since that is always present.
3. Economic uncertainty is considered for the five years only. After that, it is not allowed to increase. The rationale is that generation adequacy computations should consider economic uncertainty in the short-term (five years) when most resource plans cannot be easily adjusted for deficiencies due to generation or demand-side resources. After that time, resource plans can be adjusted to account for divergent economic paths.

An approach to represent uncertainty in generation hourly adequacy models is to adjust the 8760 hour load profile “up” and “down” to reflect uncertainty in load levels.[3] These adjustments, however, leave the order of relative values of chronological hourly loads unchanged. Additionally, such a process assumes that weather affects all geographic areas equally across the entire interconnection simultaneously. We believe that much uncertainty in the final result can not be captured by this approach because there is no variation (uncertainty) in the shape of the demand model.[4] A “family” of 8760 hourly load profiles, each reflecting different weather scenarios at distinct probabilities, may be an answer. But this family of 8760 load profiles would need the consistency among all neighboring systems described in the previous section.[5]

The illustration below shows the impact in load shape variability on loss-of load probability calculations. [6] The same weather and economic forecast error were incorporated into each load shape but there was still a significant difference in results. Each load shape was scaled to the same peak and energy so the difference between the load shapes stems from the contributions to LOLH from the second highest peak load, the third highest peak load, the fourth highest peak load and so forth.

Assistance Requested from the Load Forecasting Working Group (LFWG) and the Loss-of Load Expectation Working Group (LOLEWG)

The GTRPMTF requests that the LFWG and the LOLEWG consider how the load modeling issues described herein could be addressed in an Interconnection-wide modeling approach.

1. Since the LFWG’s purpose (from its scope) is “to assess the degree of uncertainty inherent in [the NERC aggregations of regional/subregional load forecasts], and increase consistency of forecast reporting” (emphasis added) we believe they would have the lead role in recommending approaches to address these issues, including recommending methods for aggregating independently forecasted hourly load data. This may require benchmarking to historical relationships. In the U.S., the reporting of historical hourly load data has been required by FERC Form 714. See http://www.ferc.gov/docs-filing/forms/form-714/overview.asp.
2. While the development of a family of load shapes may be desirable, the GTRPMTF understands that that may require additional work. The LFWG’s focus should first be on developing forecasted load shapes that can be aggregated in each Interconnection and each subregion in the LTRA. If forecasted load shapes are changing, that would also need to be addressed, but again, that is a lower priority at this point.
3. We would look to the LOLEWG to draw upon its members’ experience – have they recognized these issues, and, if so, how have they addressed them?

2

[1] See the GTRMPF scope at http://www.nerc.com/docs/pc/gtrpmtf/GT%20Reliability_Planning_Models_Task_Force_Scope_06-10-09.pdf.

[2] Practically speaking, we will not likely model the entire Interconnection in one model, but large parts will be modeled so that the impacts of capacity assistance from neighbors can be properly accounted for. Capacity assistance from neighbors depends not only on their generation availability, but also on their loads. Neighbors are assumed to make non-firm generation available to a system that would otherwise lose load if the assistance was not provided. This assumption is a fundamental one for all resource planners. A large amount of load diversity between neighbors (e.g., one is summer peaking while the other is winter peaking) can significantly reduce both of their capacity needs if this characteristic is incorporated into capacity planning. Therefore, load diversity should be properly represented in reliability assessments.

[3] While this is a present constraint, if other methods were developed, those would likely be incorporated by vendors.

[4] For example, maintenance scheduled for generation that is based upon a single load profile will not capture the reliability impacts of higher loads during those scheduled maintenance periods.

[5] Software vendors differ in their ability to address load forecast uncertainty. However, any solutions that developed should not be constrained by current software limitations – if a solution makes sense, vendors can adapt their models to that solution.

[6]This figure was provided courtesy of Kevin Carden of Astrape Consulting.