Day-Ahead Residential Load Forecasting based on Variational Mode Decomposition and Hybrid Deep Networks with Granger Causality Feature Selection

Accurate forecast of residential load consumption enables utilities to make informed decisions that ensure a consistent power supply. Forecasting power usage, on the other hand, is difficult owing to the nonlinear and nonstationary characteristics of time series load data that is highly correlated with weather parameters. In order to solve this issue, the paper offers a variational mode decomposition (VMD) based load forecasting technique paired with a hybrid deep learning model integrating long short-term memory (LSTM) and gated recurrent unit (GRU). VMD is a non-recursive signal processing approach that splits time series into band-limited modes referred to as intrinsic mode functions (IMFs). The IMFs are then predicted by the LSTM and GRU networks, and the results are merged to get the final forecast. The suggested model forecasts day-ahead load consumption with a resolution of 30 minutes. The results show that the VMD-LSTM-GRU model performs better than other tested recurrent models.