Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Feature-preserving reductions of high-dimensional weather forecasts for power system optimization


Lukas Schwander

Joe Warrington

Efficient real-time operation of power systems is made increasingly difficult by the growth of intermittent renewable energy sources (primarily wind and solar power). In order to forecast these power, increasingly sophisticated numerical weather prediction models are able to run multiple scenarios of wind speed and solar radiation. A group of such scenarios is referred to as an ensemble forecast.
These forecasts represent a huge amount of data that can (in principle) be used to drive a real-time power system optimization algorithm. In particular, variations between the scenarios in the ensemble forecast can be used to assess the uncertainty in the prediction of wind and solar power output. Sizing such uncertainty is critical to procuring an appropriate amount of reserves and deciding how to operate conventional generators and storage devices around the intermittent power infeeds. However, use of the full forecast data set will lead to an intractable optimization problem. We therefore need to be able to do the following:
  1. Develop approaches for reducing the dimension of the high-dimensional data set, but still capturing the main correlations (in space and time) of the forecast, in order to avoid plugging supercomputer outputs directly into an optimization routine;
  2. Develop general methods for reducing the representation of the uncertainty set surrounding a "nominal" forecast, to avoid having to include the full representation in an optimization. This should ideally provide some (probabilistic) guarantees on the safety of the system, given that the weather model only includes a limited number of scenarios;
  3. Develop novel optimization formulations exploiting as much of the above two points as possible, and test the approach on a suitable power system driven by the weather model.

This project will build on early-stage existing work on the subject, and will be aided by a colleague from the UK with access to the supercomputer outputs described above. For background on the multi-stage optimization approach previously developed, see this paper.

Mode analysis of wind power forecast errors for 336 wind farms located across Britain, derived from ensemble forecast outputs.

Weitere Informationen

John Lygeros

Art der Arbeit: 70% theory, 30% implementation
Voraussetzungen: Some optimization knowledge is essential, and interest in working with large data sets would be a plus
Anzahl StudentInnen: 1
Status: done
Projektstart: Summer/Sept 2017
Semester: Autumn 2017