The threat of climate change has major implications for planning and designing future power systems. The drastic restructuring necessary to move away from fossil-fuel generation towards renewables leads to increasing weather-sensitivity. Power system planning is therefore exposed to increasing levels of climate-based uncertainty, the detailed assessment of which remains computationally intractable insofar as it requires the combination of very large climate datasets with sophisticated power system models (PSMs). This paper therefore introduces a novel approach to overcome this challenge, enabling very large climate datasets to be efficiently subsampled such that climatologically robust power system simulations can be achieved.
In power system planning, PSMs are frequently employed. These correspond to techno-economic based computer simulations of a chosen electricity system, incorporating decisions on investment and plant operation. Recent studies indicate that, due to natural climate variability, PSM outputs (corresponding to system design parameters) depend strongly on the choice of demand & weather data (e.g. which year) considered. For this reason, robustly determining optimal power system design requires long samples of data (spanning multiple decades or centuries). Simultaneously, high-frequency (typically hourly) input weather data is required to accurately simulate the properties of variable renewables. This combination — long records at high frequency — is typically computationally unfeasible for “larger” and more sophisticated PSMs.
A new climate subsampling methodology is developed on a simple model of the United Kingdom power system. It is shown that while established timeseries reduction approaches can lead to significant errors in estimates of optimal system design and supply capacity shortages, the new methodology achieves accurate estimation of optimal system design with full generation adequacy at greatly reduced computational cost.
The new methodology implies that it is now possible to run more sophisticated power system models with very large climate datasets (e.g. multi-model multi-centennial CMIP-style GCM ensemble datasets corresponding to many millennia of weather data). This should allow a better understanding of the impacts of long-term climate variability, rare weather events and climate change, relevant to both policy-makers and industry.