In this paper machine learning models are estimated to predict electricity prices. As it is well known, these models are extremely flexible, can be used to include exogenous variables and allow to account for possible non-linear behavior of observed time series. Random forests (RF) and Support Vector Machines (SVM) are considered and their performances are compared with those of linear AutoRegressive (AR) models, with and without LASSO penalization. The application to Italian electricity spot prices (day-ahead market) with the inclusion of exogenous variables like forecast demand and wind generation and intra-day prices, has revealed that the prediction performance of the simple AR model is mostly better than the machine learning models. Only the SVM model seems to be a good competitor of the AR model, but even when its loss function is lower, the performance gain is hardly statistically significant.
Machine learning models for electricity price forecasting
Silvia Golia;
2021-01-01
Abstract
In this paper machine learning models are estimated to predict electricity prices. As it is well known, these models are extremely flexible, can be used to include exogenous variables and allow to account for possible non-linear behavior of observed time series. Random forests (RF) and Support Vector Machines (SVM) are considered and their performances are compared with those of linear AutoRegressive (AR) models, with and without LASSO penalization. The application to Italian electricity spot prices (day-ahead market) with the inclusion of exogenous variables like forecast demand and wind generation and intra-day prices, has revealed that the prediction performance of the simple AR model is mostly better than the machine learning models. Only the SVM model seems to be a good competitor of the AR model, but even when its loss function is lower, the performance gain is hardly statistically significant.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
