Planning reliable wind- and solar-based electricity systems
Key finding. Least-cost systems planned on 40 years of weather data are needed for a wind, solar, and battery system to serve all demand across a decade of previously unseen weather in half of operational tests, whereas 15 years suffices if dispatchable generation may supply 5 per cent of electricity demand.
What question did this research address?
Electricity systems have historically been planned against averaged demand and generation with a reserve margin on top. Wind and solar break that method, because their output varies on every timescale from seconds to years, and the events that threaten reliability are multi-day resource droughts rather than deviations from an average.
Capacity expansion models are typically run on one, or a few, years of weather. This paper asked how much weather history a planner actually needs — how resource adequacy improves as more years enter the planning process, what that improvement costs, and whether the cost is worth paying.
What did we find?
Forty-two years of ERA5 weather over the contiguous United States, from 1979 to 2020, were used to build wind and solar profiles and matched synthetic demand at four-hour resolution. Systems were planned to zero lost load on a sample of years, then operated on ten years of weather the planner had never seen, with capacities held fixed.
Systems planned on a single year routinely failed in operation, losing 0.082 per cent of demand in the solar-wind-battery case, 0.0074 per cent with dispatchable generation permitted, and 1.00 per cent with hydrogen storage.
The first extra year of data is the most valuable one. Adding a second planning year cut lost load by more than half in every configuration, with the largest drop, 61 per cent, in the case with dispatchable generation.
Reaching complete resource adequacy in half the decade-long operational tests took 15 planning years with dispatchable generation available, and 40 planning years for both the solar-wind-battery and the hydrogen configurations.
More planning years also change what gets built, not merely how much. Between one and forty planning years, solar capacity rose 71 per cent and battery capacity 110 per cent in the solar-wind-battery case, while wind capacity fell 18 per cent — severe multi-day wind droughts are far more common in the record than severe solar droughts, so a longer sample is more likely to contain one.
A small amount of dispatchable generation does most of what decades of data do. Limited to 5 per cent of annual demand but running in about 20 per cent of hours, it held the increase in levelized cost between one and forty planning years to 3.0 per cent, and left asset capacities largely insensitive to which weather years were sampled.
Dispatchable generation also shrinks the system. Planned on one year, solar-wind-battery systems carried 43 per cent more solar and 37 per cent more wind capacity than systems with dispatchable generation, and curtailed 60 per cent of available wind and solar output against 38 per cent.
In the hydrogen configuration, more planning years mainly bought long-duration storage — hydrogen storage up 62 per cent and fuel cell capacity up 45 per cent, while battery capacity fell 41 per cent and wind, solar, and electrolyzer capacities each moved less than 15 per cent.
Reliability bought this way is not always worth its price. Valuing lost load at $10 per kilowatt-hour, the marginal cost of avoided lost load exceeded that value beyond one planning year for solar-wind-battery systems and beyond two with dispatchable generation. For hydrogen systems it stayed below $10 per kilowatt-hour all the way to 40 planning years.
The worst operational failures shared a signature — severe wind droughts with 24-hour average output near 10 per cent of nameplate capacity, in any season, and worst from April through October when high solar output coincides with high cooling demand.
Results were essentially unchanged at one-, two-, three-, and four-hour time steps, because the events that drive system builds last many hours or longer.
Why does it matter?
It puts a number on a gap between research practice and reliability. Single-year and few-year planning studies are common, and this shows that systems designed that way lose load when the weather does something the sample did not contain.
The result is not simply that reliability is expensive — it is that the last increments of reliability bought through weather data are expensive, while the first are cheap. One extra planning year halves lost load; the marginal cost then climbs past any reasonable value of lost load within a few more years.
Five per cent dispatchable generation substitutes for twenty-five years of weather history. That reframes the role of a small firm fraction in a renewable system as a hedge against planning uncertainty as much as a hedge against real-time shortfall.
Citation
Tyler H. Ruggles, Edgar Virgüez, Natasha Reich, Jacqueline Dowling, Hannah Bloomfield, Enrico G. A. Antonini, Steven J. Davis, Nathan S. Lewis, and Ken Caldeira (2024). Planning reliable wind- and solar-based electricity systems. Advances in Applied Energy 15, 100185.
Related
- Can wind and solar power reliably meet electricity demand?
- What does a reliable electricity system built on wind and solar actually need?
- Identification of reliable locations for wind power generation through a global analysis of wind droughts (Antonini et al., 2024)
- Geophysical constraints on the reliability of solar and wind power worldwide (Tong et al., 2021)
- Role of long-duration energy storage in variable renewable electricity systems (Dowling et al., 2020)
- Geophysical constraints on the reliability of solar and wind power in the United States (Shaner et al., 2018)