Fakultät für Mathematik und Naturwissenschaften

Masters Thesis Applying Multivariate Statistical Methods for Forecasting Electricity Price Contributors

Masters Thesis - Applying Multivariate Statistical Methods for Forecasting Electricity Price Contributors

Partner:
Universität Wuppertal Lehrstuhl Stochastik, Vattenfall Energy Trading GmbH

Task:
Price contributors to be considered:

  • Import/Export (cross-border) flow of electricity in Germany
  • Margins/add-ons

The task is to come up with better price contributor models for Germany. The approach to be investigated is a reference day/hour method. This approach is to be investigated because we think it can give better results than we have, and we have never made a thorough investigation of this approach before.

This approach is then to be compared with the present models; Principal Component Regression, PCR for margins, and Partial Least Squares regression for import/export short-term forecast and weekly mean regression f or import/export long-term forecast.

Reference day model
The aim of this method is to find historically similar days/hours by studying factors (demand, wind, etc), that affect the price contributors, and then the value(s) of the fundamentally equal day /hour or days /hours is reused.

The steps to be undertaken can roughly be explained as follows:

  1. Find a suitable set of statistically significant underlying factors (such as wind production, demand, temperature, holidays, etc) which can explain historical price contributors well enough.
  2. Investigate robust method of comparing daily/hourly price contributor to each other. Practically, single hours and single days are most likely to be compared. Linear and non-linear methods can be taken into consideration. For example discriminant analysis and Classification and Regression Trees (CART).
  3. Estimate future price contributors using the methods from point 1-2 above, by comparing the future predictors (such as wind production, demand, temperature, holidays, etc) to their past values and thereby finding the most similar historic/observed price contributor.
  4. The forecast results are then compared with the results from the present models.

Kooperationspartner:
Vattenfall
www.vattenfall.de
Industry Supervisor: E. S.

Master Candidate: Mr. L. A.

Ansprechpartner:
Univ. Supervisor: Univ. Prof. Dr. Barbara Rüdiger-Mastandrea
E-Mail: ruediger[at]uni-wuppertal.de
www.math.uni-wuppertal.de/~ruediger
 

Vorarbeiten:
Vortrag:
Datum: 18.10.2010
Zeit: 14Uhr

Referenzen:

  • Wolfgang Härdle, Leopold Simar: Applied Multivariate Statistical Analysis (Second Edition). Springer 2003, 2007
  • Dallas E. Johnson: Applied Multivariate Methods for Data Analysts. Duxbury Press, 1998
  • Rafal Weron: Modeling and Forecasting Electricity Loads and Prices. John Wiley & Sons Ltd., 2006
  • Derek W. Bunn: Modelling Prices in Competitive Electriciy Markets. John Wiley & Sons Ltd., 2004
  • Markus Burger, Bernhard Graeber and Gero Schindlmayr: Managing energy risk. John Wiley & Sons Ltd., 2007

Masters Thesis
Applying Multivariate Statistical Methods for Forecasting Electricity Price Contributors

Weitere Infos über #UniWuppertal: