A Novel Spatial-Temporal Deep Neural Network for Electricity Price Forecasting

Electricity price forecasting is a difficult task because it depends on various factors such as weather, fuel, load, and bidding strategies. These characteristics bring a lot of volatility to electricity prices. In addition, there exist coupling relationships between different price zones in Europe. CNN-based or LSTM-based methods cannot capture the relationship by their structures or there is a need to extract these couplings explicitly manually. In this work, an end-to-end graph neural network is proposed for the first time to learn the coupling between different price zones automatically. The proposed model mainly consists of two parts: a graph learning module and a temporal learning module, which both are designed to learn spatial information of different price zones and temporal information of historical data, respectively. The performance of the proposed model is evaluated on one-year public data collected from the Nord Pool. The results indicate that our model provides a better solution for electricity price forecasting.