Deep Learning Based Electrical Load Forecasting Using Temporal Fusion Transformer and Trend-Seasonal Decomposition

Modern smart‑grid operation and planning need very accurate load forecasts, both for single customers and for groups of customers, to keep the system stable and to distribute power efficiently. This project offers a step‑by‑step framework that uses deep learning. It combines the Temporal Fusion Transformer (TFT) with a trend‑seasonal decomposition method. The framework first applies a moving‑average technique to split raw, noisy load time‑series data into two parts: a long‑term trend and a repeating seasonal pattern. This split makes the model easier to understand and improves its predictions. The TFT model forms the core of the system. Its multi‑horizon forecasting ability and self‑attention layers capture complex time dynamics and how

different variables interact. To test the approach, we run simulations with historical load data, such as the regional grid data from Iran. We measure performance with standard metrics: Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE). The project provides detailed guidance for applying advanced transformer‑based models to power‑system state estimation and demand‑side management.

Message: Hi FE, I need help with “Deep Learning Based Electrical Load Forecasting Using Temporal Fusion Transformer and Trend-Seasonal Decomposition” in “Electrical Engineering”