Probabilistyczne prognozowanie cen energii elektrycznej: Ewaluacja prognoz 
wielowymiarowych i finansowa wartość prognozy
[Probabilistic electricity price forecasting: Multivariate forecast evaluation 
and the value of a forecast]

Tim Janke (Energy Information Networks and Systems Lab, TU Darmstadt, Germany) 


The paradigm in electricity price forecasting, as in many other forecasting domains, 
has shifted from point to probabilistic forecasting. Probabilistic forecasts enable 
the use of stochastic programming techniques to improve decision making under 
uncertainty, e.g., for power plant dispatch decisions. However, probabilistic 
forecasts usually come in the form of marginal distributions over the individual 
prices while actually scenarios representing realizations of the multivariate price 
vector are needed. In this talk I will first present a quantile regression neural 
network for the simultaneous estimation of the quantiles of the marginal price 
distributions. I will then show a way to generate scenarios from the obtained 
marginals using a Gaussian copula and discuss the evaluation of multivariate scenario 
forecasts. Finally, I will present an exemplary power plant dispatch problem to 
investigate the monetary value of probabilistic forecasts.