Reestymacja parametrów sieci neuronowej w prognozowaniu cen energii elektrycznej: porównanie podejść
[Everything you always wanted to know about deep learning (but were afraid to ask)]

Grzegorz Marcjasz (ORBI, PWr, Wrocław)
Andrzej Puć (Faculty of Pure and Applied Mathematics, PWr, Wrocław)


The talk introduces the problem of periodic reestimation of model parameters in electricity price forecasting. 
Using the data from two electricity markets, we show that a frequent reestimation of the parameters is key in 
achieving a good predictive performance. However, it imposes a significant computational burden. We also propose 
an alternative approach based on a partial recalibration.