Blog May 15, 2020

On Friday, May 8, 2020, the ERCOT grid observed increasingly prevalent types of market volatility, highlighting some of the issues and opportunities surrounding high renewable penetration. Following record wind generation Thursday evening (around 21 GW hourly average), declining wind forecasts remained high enough for Friday such that a low percentage of gas plants stayed online – even with cheap fuel costs. Heat indices and HVAC cooling demand across the grid were mild, especially in Dallas and the Permian Basin; the latter region has also observed reduced power usage from crude oil facilities. Consequently, Friday net load forecasts (demand minus wind minus solar) suggested wide surplus and cheap Day-Ahead/Real-Time pricing. Statistical models built on historical training data likely projected a narrow confidence band surrounding this pricing outlook.

On the contrary, our models stated high risk around Real-Time pricing for three main reasons:

  1. Constraints surrounding transmission lines transporting wind power
  2. Ramping shortages from generators needed to offset declining wind power
  3. Higher accuracy windy forecasts compared to ISO projections. 

Lo and behold, Houston and South Hubs settled at an exorbitant $212/MWh and $108/MWh, respectively. 

Figure 1: Settlement pricing at Houston (blue) and South (black) Hubs from May 8. Prices spiked highest in the early afternoon as net load climbed >10 GW in less than five hours. Source: ERCOT
Figure 1: Settlement pricing at Houston (blue) and South (black) Hubs from May 8. Prices spiked highest in the early afternoon as net load climbed >10 GW in less than five hours. Source: ERCOT

Transmission Constraints

From a big picture and traditional scope, it was accurate to assess wide surplus and minimal scarcity threat when forecasting grid conditions on Friday. ERCOT has an energy-only market design that opts for a real-time scarcity pricing mechanism in lieu of a forward capacity auction. The scarcity price mechanism, defined by the Operating Reserve Demand Curve (ORDC), applies a roughly exponential price adder across the entire systems once available surplus and ancillary resources fall below a certain threshold. However, the reserve levels used to determine scarcity adders include generators instructed to reduce output to relieve transmission constraints. In this case, capacity available for economic dispatch (i.e. surplus), was sufficient on paper and ancillaries were not deployed – though clearly there were regional shortages.

Figure 2: Capacity available to SCED on May 8 with backcast estimates for wind and solar curtailments using Genscape’s dispatch model and Bid Data API. Source: ERCOT and Genscape
Figure 2: Capacity available to SCED on May 8 with backcast estimates for wind and solar curtailments using Genscape’s dispatch model and Bid Data API. Source: ERCOT and Genscape

Numerous transmission lines threatened overloads last week. Wind production hovered near all-time highs, line maintenance limited export pathways for wind, and load imbalances grew between Dallas and a significantly hotter Houston. To limit wind power and prevent overloads, ERCOT implemented congestion costs to depress West and even some North prices near $0/MWh (standard offer for wind farms without production tax credit). Most generators submit a set of offers at different output levels and curtail output when their local price falls beneath a given offer tier. While curtailing wind solved constraints near these remotely located farms, a byproduct was lower dispatch from nearby gas and coal plants as net load accelerated. Limitations around West and North supply heightened the underlying system energy cost and fed transmission constraints to surge prices in the lower half of the footprint.

Figure 3: Shift factors for JK_TOKSW_1 345kV transmission constraint with highest congestion rent on May 8. Source: Genscape’s CongestionIQ platform
Figure 3: Shift factors for JK_TOKSW_1 345kV transmission constraint with highest congestion rent on May 8. Source: Genscape’s CongestionIQ platform

Ramp Constraint

Weak price signals delivered to West and North zones, including the most liquid for over-the-counter settlement point, North Hub, restricted generation response to gas and coal plants within the South and Houston zones. Even without transmission constraints, increasing output by around 11 GW over five hours is a daunting task for a thermal supply stack largely engineered decades ago for continuous, constant runs. Wind penetrating 45-50% of demand as late as 11:00 AM also means that many thermal plants are starting from minimum values or potentially offline entirely. Adding hefty transmission limitation only exacerbates the ramp constraint.

Each generation resource has a high sustainable limit, or ceiling, where it can produce. When summing ceilings of all available/online resources, we obtain total capacity used to deduce surplus. Yet, instead of the overall capacity totals, our analysts focused on high dispatch limits for resources that assess the max generation level a resource can achieve within five minutes. Telemetered ramp-rate data from a windy day this past December in Figure 4 below highlights early exhaustion of gas generation as wind declines and evening demand builds. Many of the quickest generators, namely gas turbine peakers and batteries, commit primarily in the ancillary market and could stay offline if ORDC is not enabled – as was the case on May 8. As thermal plants struggled to respond, average marginal energy costs soared and scaled costs of the underlying transmission constraints, creating a positive feedback loop.

Figure 4: Cumulative ramp-up rates of supply stack on December 5, 2019, where wind levels underperformed forecast by several GW. Data source: ERCOT
Figure 4: Cumulative ramp-up rates of supply stack on December 5, 2019, where wind levels underperformed forecast by several GW. Data source: ERCOT

Wind Generation Forecast

Arguably the largest warning sign for real-time pricing threat stemmed from a stellar wind forecast from our head meteorologist, Jefferson Rhoads. Based on a neural network model, various weather vendors and sources, and economic curtailment projections, our ~2.5 GW weaker wind forecast throughout the afternoon sounded alarm bells for insufficient unit commitment in the Day-Ahead solution. Both ERCOT and Genscape forecasts presented steep declines, which magnifies error in a relatively coarse execution interval of the Day-Ahead solution (hourly) that prepares for instantaneous wind intermittency. While we don’t explicitly forecast solar within ERCOT, congestion forecasts and cloud-cover projections were communicated to clients before solar ended up underperforming by more than 50%. As solar increases by three times and wind climbs at least another four GW by 2022, economic curtailments as observed on May 8 will only become more consistent before 345 kV transmission projects are energized.

Transmission
Figure 5: Genscape and ERCOT Day-Ahead wind forecasts versus actual output. Source: Genscape and ERCOT
Figure 5: Top: Genscape and ERCOT Day-Ahead wind forecasts versus actual output. Bottom: ERCOT’s solar forecast versus actual output. Source: Genscape and ERCOT
Figure 6: ERCOT’s solar forecast versus actual output. Source: ERCOT

Key Takeaways

In an era of rapidly increasing renewable penetration, ERCOT will continue to observe volatility transform. Reserve margins are beginning to expand, and to some degree, quell scarcity threats around winter and summer peaks. However, new threats are dispersed year-round: sharp variations in supply, inadequate transmission infrastructure (including stability concerns around pockets of asynchronous generators), limited storage, outdated thermal plant technology, and greater influence in forecast errors. At Genscape, our services help navigate this evolution for participants on the front lines and beyond. To learn more about our offerings or to request a demo, please click here.