Forecasting Solar Energy Production using a Hybrid GCN-BiLSTM Model

Abstract

Under increasing levels of renewable energy source (RES) penetration, unpredictability and uncertainty are emerging drivers of power imbalances. Forecasting is frequently used to anticipate renewable energy power generation. Forecast errors, on the other hand, significantly negatively impact power system performance. This research describes a deep learning technique based on spatiotemporal analysis for accurately forecasting solar power generation. Solar power generation output from seven PV sites is predicted using a hybrid graph convolutional network (GCN) module, bidirectional long short-term memory (BiLSTM) module, and attention layer. Our model effectively captures comprehensive spatiotemporal correlations on real-world solar power generation datasets and surpasses several existing methods.