Deep Learning Based Time-Series Forecasting

Abstract: Time-series forecasting plays a vital role in domains such as energy management, finance, healthcare, and smart infrastructure. This paper presents a deep learning–based framework for short-term electricity consumption forecasting using historical power usage data. The proposed system evaluates and compares three advanced architectures: Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Transformer-based models. A multivariate forecasting approach is adopted by incorporating power-related and calendar-based features. A sliding window strategy is used to predict future values from recent observations. The system is deployed through a Flask-based web interface that enables interactive visualization and real-time prediction. Experimental results indicate that the CNN model achieves the best performance for short-term forecasting, outperforming LSTM and Transformer models in terms of RMSE, MAE, and R² score. The proposed framework demonstrates the practical applicability of deep learning techniques for intelligent energy demand prediction.

Keywords: Time-Series Forecasting, Deep Learning, LSTM, CNN, Transformer, Energy Consumption, Flask