Resilience and extreme events in energy system models - best practices for stress tests and datasets

Workshop
, , , ,
1
  • genre: break‑out‑group
  • title: Resilience and extreme events in energy system models - best practices for stress tests and datasets
  • presenter: @aleks-g, @luk23
  • description:

We suggest a break-out group in which we can discuss common approaches and practices to resilience and extreme events. Hopefully we can compile a document that could be useful to the openmod community, which we could share afterwards or upload in the wiki.

We want to go through some fundamentals: how do we define resilience for energy modelling? How do we implement such considerations from a more qualitative, planning perspective and how from a quantitative modelling perspective?

Are there any existing frameworks and open data that can be helpful as guidance for stress tests or scenario generation? How do we study these matters systematically and not just as a collection of all possible scenarios we could come up with?

Lukas and I are happy to give more input on how this can be done in the context of weather and climate resilience and extreme weather. But also here, we need to agree on what falls under this category: natural disasters, infrastructure damage, dunkelflauten, climate change.

What climate and weather data can we use to study and answer such questions? Usually, energy system models use reanalysis data (such as ERA5), but synthetic data, climate models (e.g. as part of CMIP6), expired weather forecasts could increase the number of realisations and thus improve the assessment of risks. Which open tools and packages can we use to study rare, but extreme situations in the energy system?

One objective of this break-out session would be to coordinate and harmonise different approaches that have been used across different open models and exchange insights.

  • background: Craig et al, 2022 and McCollum et al, 2020. @luk23 is a PhD student at the Technical University of Denmark. @aleks-g is a postdoc (also at DTU). Both use PyPSA-Eur to investigate the impact of extreme weather and climate change on highly renewable energy systems.
6 Likes
2

Here’s a summary of what we discussed during the break-out session:

Definition:

  • Context-dependent; broadly, the capacity of a modelled system to withstand or to recover quickly from difficult situations
  • Similar to sustainability in its vagueness
  • Resilience in practice should be associated with a risk factor / challenge, i.e. we should ask: Resilient against what?
  • Goal is to predict (cost of) operations, limit occurrence/return times of certain events or stress test the system.

Use cases and risk factors:

  • often connected to uncertainties; usually uncertainties relate more to the input side of the model, whereas resilience is desired/tested on the outputs
  • need to distinguish between unforeseen situations or scenarios (unlikely, not considered, black swans) and unforeseeable situations (inherently uncertain such as weather)
  • can be of techno-economic nature (costs, sector coupling, technology availability), socio-political nature (conflicts, war, migration, human behaviour, market structures, demographics) or environmental (weather, natural disasters, climate change)
  • exists on time scales from seconds (grid stability/blackouts) to hours (resource availability) to weeks (dunkelflauten) to years and decades (large trends)

Connection to modelling:

  • Questions might differ by region (global south v global north); resilience might be an even bigger challenge for African countries with growing populations and scarce existing infrastructure; resilience against conflicts and instability has been less of a concern in Europe (until now)
  • Need to evaluate whether ESOMs and their techno-economic focus are always the right tool. Computationally prohibitive to run all imaginable scenarios.
  • Qualitative approaches that focus on storylines and explore them can avoid the fallacy of using one single objective (risks might not be quantifiable)
  • Quantitative approaches can include adding additional constraints (on infrastructure/costs/resource availability/spatial distribution of infrastructure), multi-objective optimisation, scenario analysis, stochastic and robust programming, or near-optimal solutions/MGA

Best practices:

  • Storylines can help in setting up scenarios and identify possible compound events (and cascading effects) whose economic impacts can be studied.
  • Be uncomfortable with uncertainty and do not assume probabilities we do not have; clear communication is key and representing blind spots in a suboptimal way is better than maintaining blind spots.
  • Ask yourself whether we want to test resilience on an existing system or whether we want to include resilience in the planning / optimisation.
  • Perturbations, stress tests, and out-of-sample analyses can be useful.
  • Near-optimal spaces / modelling-to-generate alternatives can be helpful to explore alternative, more robust solutions, but also by validating claims when relaxing constraints.
  • Shadow prices / dual variables can identify sensitive constraints or variables.
  • Talk to colleagues from other disciplines, they might deal differently with uncertainties (e.g. climate science) and have a better understanding of risk factors and their consequences.
  • Consider using myopic foresight or rolling horizons instead of perfect foresight.
  • Define what kind of resilience you are looking at, and how this informs your motivating, choice of metrics, and acknowledge trade-offs between scenarios and computational efforts.