Dunkelflaute

The argument

Average renewable generation is a poor description of reliability. A defensible system design names the low-generation event it is built to survive, shows how it survives it, and states the event that would still defeat it.

Turbines, panels, batteries, transmission, reserve plant and forecasts appear in this course only where they change how that event is defined, detected, predicted, survived, allocated or bounded. It is not a survey of renewable technology. It is one problem, examined for twelve weeks.

The course argues from a single synthetic fortnight, so every week is talking about the same weather. Here it is.

Mean output across the fortnight

27% of capacity Synthetic trace

Hours below the 12% threshold

88 of 336 Synthetic trace

Longest qualifying event

61 hours Synthetic trace

Both descriptions are true of the same system. Only one of them is a design input. Which figures here are the course's own chosen boundaries and which are properties of the generated trace is marked everywhere they appear — see the event for the definition all of it resolves against.

The semester

Three acts, twelve claims. Each week owns exactly one property of the event, and no two weeks own the same one.

Act IWeeks 1–4

What the event isVisibility, definition, correlation and spatial extent. The act that decides what the rest of the course is allowed to see.

Act IIWeeks 5–8

What survives itDepth below cut-in, the hours when the wind–solar complement fails, duration, and the economics of an asset you hope stays idle.

Act IIIWeeks 9–12

Who decides, and who explains it afterwardsAllocation, forecast lead time, the institutional shape of a post-mortem, and a system you are willing to defend.

Five weeks worth arriving for

1. W01The Average Is Lying to YouTwo numbers describe one system, and only one of them is a design input.
2. W02When Does Bad Weather Become a Drought?Who decides where the line sits, and what does moving it discard?
3. W06Solar Helps, Until It Doesn'tComplementarity holds on average — here is the week it stops holding.
4. W07Storage Is Two Problems Wearing One NameDuration, not power, turns out to be the constraint that binds.
5. W12The System You Are Willing to DefendYou state the event that defeats your design, out loud, to a room.

All twelve weeks →

Assessment

Four stages of one problem. Each stage consumes what the last one produced, so the semester assembles a single defended design rather than four unrelated submissions.

1. A1 Define the Drought

   Takes as input

   —

   Produces

   an operative definition of the event

   Weight

   15% of the course

2. A2 Site Against Failure

   Takes as input

   A1's definition

   Produces

   a portfolio with a correlation profile

   Weight

   20% of the course

3. A3 The 72-Hour Dispatch

   Takes as input

   A2's portfolio

   Produces

   a dispatch log and a counterfactual

   Weight

   25% of the course

4. A4 Design for the Worst Week

   Takes as input

   A3's dispatch log

   Produces

   a system, and the event that beats it

   Weight

   40% of the course

Full assessment detail →

Who it is for

Postgraduate students from engineering, computing, data science, environmental science and public policy. No power-systems background is assumed and no machine equations are derived. If you can read a time-series chart and argue about what it means, you can do this course.

By week 12 you can take a reliability claim apart: name the reporting convention behind it and what that convention cannot see; then state, in threshold and duration, the low-generation event a design survives; run that design hour by hour through the event; and name the hour it stopped being recoverable. You will also be able to say what would defeat it — in numbers someone else can re-run.

The weekly pattern is a two-hour seminar followed by a two-hour studio. The studio is where the work happens.

Where to go next