Short- and long-term forecasting of electricity prices using embedding of calendar information in neural networks

Electricity prices strongly depend on seasonality of different time scales, therefore any forecasting electricity has to account for it. Neural networks have proven successful in short-term price-forecasting, but complicated architectures like LSTM are used integrate the seasonal behavior. This paper shows that simple neural network DNNs with an embedding layer information can generate a competitive forecast. The embedding-based processing calendar additionally opens up new applications trading, such as generation price forward curves. Besides theoretical foundation, this also provides empirical multi-year study German market both and derives economical insights from layer. price-forecasting mean absolute error proposed is better than time-series benchmark models even slightly our best model sophisticated hyperparameter optimization. results aresupported by statistical analysis using Friedman Holm’s tests. • embeddings Generation hourly curves Case-study