Automated deep learning and Internet of Things framework for building energy management: A university case study

Abstract

Monitoring energy consumption in buildings presents significant opportunities, especially in developing economies like India. However, current solutions often overlook cost-effective, small-scale, accurate, and open-source data-driven methodologies. Research in this area is often hindered by concerns related to security and privacy, high investment costs, and unpredictable returns. To address these challenges, we developed an automated hybrid deep learning and Internet of Things (DL-IoT) building energy management system (BEMS) aimed at conserving energy. The DL-IoT combines deep learning techniques with fuzzy logic to effectively manage uncertainty and noise in electrical properties. Our DL-IoT regression model demonstrated low mean absolute error and mean squared error, achieving a coefficient of determination of 0.99 for out-of-sample energy consumption predictions. We extracted twenty-seven electricity usage variables from raw data to train the model. Experimental results revealed a linear relationship between these characteristics and energy use. The proposed model successfully predicted features that could contribute to energy savings, such as Power Factor and Power in the Y Phase. Specifically, it estimated that a one-unit increase in Power in the Y Phase and Power Factor would result in a reduction in energy consumption. The findings of the experiment indicated that the model captured the variability of the data better than other models. The results demonstrated the superiority of the proposed model over other mainstream existing models. Through the results of this paper, a more efficient energy data management and consumption plan can be established.