Deep learning based smart energy consumption prediction in residential buildings

The prediction of building energy usage is crucial for improving decision-making and energy conservation in the construction industry. Accurate forecasts of energy consumption enhance resource optimization, particularly in smart cities where interconnected devices manage operations autonomously. These devices consume substantial energy, making energy optimization essential. The advent of smart homes has heightened the need for intelligent applications in healthcare, asset management, security, automation, and energy management to understand residents’ behaviors and predict future demands. Optimizing energy grids and power stations is necessary to minimize environmental impacts. Additionally, creating energy-efficient buildings can significantly reduce overall energy consumption, also contributing to lower carbon emissions, reducing energy waste, and overall ecological sustainability. Deep Learning (DL) methods, particularly Bidirectional Long Short-Term Memory (BLSTM) networks, are recognized for their effectiveness in prediction tasks, including energy consumption forecasting. Predicting energy demand is crucial as smart cities continue to integrate advanced technologies for efficient resource management. Many research studies focused on monthly or annual prediction. This research employs a B-LSTM network to forecast household energy consumption using hourly, daily, weekly, and monthly data, especially within smart homes, which generalizes across multiple time scales. The Individual Household Electric Power Consumption (IHEPC) dataset is a diverse and large collection of energy consumption data from smart homes, for testing predictive models. The proposed Energy Consumption Prediction models including extreme gradient boosting (XGBoost), categorial boosting (CatBoost), Gradient Boosting, BLSTM used in this study and demonstrated a lower error rate compared to previous approaches. The findings of this study are valuable for policymakers and leaders to make more informed energy investment decisions. Future work will explore the scalability of the model for larger and more diverse datasets.