Forecast of Energy Demand Using Temporal Fusion Transformer

This research paper presents an application of Temporal Fusion Transformer (TFT) model for time series forecasting using deep learning techniques. The novelty of TFT model lies in combining the strengths of recurrent and convolutional neural networks, ultimately resulting in improved accuracy in forecasting. The dataset used in this study consists of data on power consumption per quarter hour of 370 consumers, over a four-year period. The results show that the TFT model outperformed traditional time series models and achieved a lower Mean Absolute Error (MAE) and a lower Root Mean Squared Error (RMSE) in predicting future energy consumption. The paper indicates that the TFT model can be an effective tool for accurate and reliable time series forecasting in various industries, including energy and finance.