The growing demand for intelligent energy use requires systems capable of predicting consumption behavior in real time and adapting to different operational environments. Traditional forecasting methods often lack flexibility when integrated into modern energy monitoring platforms. Advances in neural network architectures offer alternatives for capturing nonlinear and dynamic consumption patterns. Energy forecasting also plays a central role in optimizing distributed systems and reducing operational uncertainty in energy management. This study introduces an intelligent software system designed to perform real-time energy consumption forecasting, integrated with Energy Management Systems (EMSs). The proposed solution communicates with sensing devices via the MQTT protocol, allowing continuous data acquisition and flexible system integration. Two forecasting models were implemented: a hybrid ARIMAX-NN model that combines statistical methods with neural networks and a CNN-LSTM Autoencoder (CNN-LSTM-AE) model that captures temporal dependencies and nonlinear behaviors. Public datasets from residential and commercial buildings were used for model validation. The software adapts to different input configurations without requiring structural changes, supporting a wide range of metering devices and data formats. Forecast results are updated in real time and can be seamlessly integrated into operational environments. The system's modular design enables future expansions such as graphical interfaces and alert generation mechanisms. This approach provides a scalable foundation for supporting energy efficiency initiatives in residential, industrial, and commercial applications.