Fourier Transform Temporal Features based Hybrid GRU-TCN Algorithm Performance Analysis for Prediction of Multi-Region Electricity Energy Consumption

Energy Consumption forecasting is an important part of proper management and sustainability for the grid. Deep learning models are currently being employed as strong energy demand predictors. Here, Simple Imputer is used to fill in missing data, and LOF is used to remove the outliers in this work. The temporal features extracted by the Fourier Transform are then fed to models LightGBM and GRU+TCN for evaluation. Experimental results show that the GRU+TCN model outperformed the PJMW region with the lowest MSE of 0.0173 and R² of 0.9699, revealing more accurate prediction capability.