Transfer learning for electricity price forecasting

Electricity price forecasting is an essential task in all the deregulated markets of world. The accurate prediction day-ahead electricity prices active research field and available data from various can be used as input for forecasting. A collection models have been proposed this task, but fundamental question on how to use big often neglected. In paper, we propose transfer learning a tool utilizing information other We pre-train neural network model source finally do fine-tuning target market. Moreover, test different ways rich forecast 24 steps ahead hourly frequency. Our experiments four indicate that improves performance statistically significant manner. Furthermore, compare our results with state-of-the-art methods rolling window scheme demonstrate approach. method algorithms by 7% French market 3% German