Forecasting Electricity Spot Market Prices Based on LSTM and GCN

Forecasting prices of spot electricity markets is crucial for power system operation and market decisions. This paper proposes a hybrid model combining Long Short-Term Memory (LSTM) and Graph Convolutional Network (GCN) to address the spatial-temporal complexity in forecasting electricity prices. The model uses LSTM to capture temporal features at individual grid nodes and GCN to model spatial correlations based on grid topology. Through end-to-end training, the model integrates temporal and spatial factors effectively. Key innovations include spatial-temporal coupling, dynamic adaptation to grid changes, and enhanced interpretability for forecasting locational marginal prices. Potential improvements like multi-head attention mechanisms are also discussed, highlighting potential of the model for accurate and reliable price forecasting in dynamic electricity markets.