SMART ENERGY MANAGEMENT IN EDUCATIONAL INSTITUTIONS AN IOT-ENABLED SYSTEM FOR REDUCING ELECTRICITY CONSUMPTION AND CARBON EMISSIONS

Abstract

Educational institutions face increasing pressure to reduce energy consumption, operating costs, and carbon emissions, while simultaneously satisfying acceptable indoor conditions. In this study, the analytical component of an IoT-enabled smart energy management framework was developed and evaluated based on the electricity-meter data, weather data, temporal features and building characteristics of educational buildings. Following screening and preprocessing, 604,381 observations from 36 buildings were retained, with a mean building-level data completeness of 97.64%. The five forecasting models were compared: persistence, Multiple Linear Regression, Random Forest, XGBoost, and Long Short-Term Memory. XGBoost had the best results with an MAE of 7.50 kWh, RMSE of 12.80 kWh, NRMSE of 13.20%, RMSLE of 0.17, and R2of of 0.91. The most significant variables were lagged electricity consumption variables, hour of day, outdoor temperature and building floor area. A residual analysis was performed on each building using a 95th percentile threshold to identify candidate excess-consumption events, resulting in 205 events. These events were observed in 5 selected buildings and represented 119,900 kWh or 11.99% of the total consumption. The corresponding potential financial and carbon reductions were ₹959,200 and 83.93 tonnes CO₂e, respectively. The results demonstrate the ability of machine-learning forecasting and residual analysis to detect unusual demand and support future IoT-enabled monitoring and control.