<p>Household energy consumption is a phenomenon that should be predicted with great accuracy, especially in places with high weather variability, where weather conditions are crucial determinants of energy demand. Conventional forecasting models fail to adequately account for the effects of weather-related factors, and hence, they cause poor management of energy. This work analyzes the results of using the Temporal Fusion Transformer (TFT) model to predict household energy consumption in two situations: with and without weather parameters (temperature, humidity, and pressure). A real-world dataset is used to apply the model to the time period between November 5, 2022, and January 5, 2024, and evaluated using various time scales, such as minute, day, week, and month predictions. The findings prove that the use of weather parameters, especially temperature, can greatly enhance the prediction accuracy. TFT model achieves superior performance consistently in comparison to the benchmark deep learning models, such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (BiLSTM), as well as Meta Graph-based Convolutional Recurrent Network (MegaCRN), across all time scales. Moreover, the analysis shows that the most significant impact on energy consumption is produced by temperature, and combined weather effects also give further predictive information. All in all, the results indicate the significance of using meteorological data and advanced deep learning models to improve the accuracy and reliability of energy consumption forecasting to make energy management systems more efficient and sustainable.</p>