Short-term Spatiotemporal Load Forecasting for Norwegian Bidding Zones

Abstract

Short-term load forecasting is vital for electric utility companies. The objective of this thesis is the short-term load forecasting of the five bidding zones in the Norwegian electrical grid. This master thesis proposes a novel method of approaching short-term load forecasting problem called Lagged SpatioTemporal Features Short-Term Load Forecasting(LSTF STLF) using LSTM. LSTF STLF is based on a spatiotemporal feature selection approach. The dependencies between the five of Norway's bidding zones in the Nord Pool power market are discovered using tools such as correlation and mutual information to find the best spatiotemporal features from all bidding zones to better perform electricity demand forecasting in each given zone. By applying the proposed spatiotemporal feature extraction approach, forecasting accuracy improved significantly for the five bidding zones on a 48-hour forecasting horizon.